CN101320387A - Web page text and image ranking method based on user attention time - Google Patents

Abstract

The invention discloses a web page text and image sorting method based on user attention time. The method includes the following steps: 1) using the attention time to personalize the existing web page sorting, so that the sorting results conform to the user's psychology; 2) using a customized browser to collect sample information of text attention time; 3) using a customized browsing 4) Correct the collected attention time samples; 5) Predict the attention time of unknown web pages based on the similarity between text and pictures; 6) Use the attention time combined with traditional search technology to generate personalized Pages and pictures are sorted. The invention effectively combines the user's preferences in the search process, so that the final ranking result is closer to the ideal ranking expected by the user, so that the web page text and image search engine provides better personalized services for the user.

Description

Web page text and image ranking method based on user attention time

technical field

The invention relates to the field of computer search, in particular to a method for sorting web page text and images based on user attention time.

Background technique

Existing personalization engines rely on user feedback, which can be divided into explicit feedback and implicit feedback. From both kinds of feedback, we can obtain user preference characteristics (Salton & Buckley 1990; White, Jose, & Ruthven 2001; White, Ruthven, & Jose 2002). But users are generally unwilling to provide explicit feedback, so more and more researches now turn to implicit feedback (Granka, Joachims, & Gay 2004; Guan & Cutrell 2007; Fu 2007). Studies have shown that implicit feedback can well reflect the user's search intention (Fox et al.2005; Dou, Song, & Wen 2007; Fu 2007). And user preferences obtained from a large number of implicit feedback are often better than explicit feedback more reliable. Query history: In modern research, the most frequently used implicit feedback is the user's query history. Google's personalized search (http://www.google.com/psearch) is based on the user's query history. In general, algorithms based on query history can be divided into the following two categories: one is based on the entire query history, and the other is based on a certain query session (referring to a series of related queries). For the former, usually the algorithm will generate a profile document of the user to describe the user's search preferences. Click data: Click data is another very important type of implicit feedback, such as (Dupret, Mrudock, & Piwowarski 2007; Joachims 2002). On a search results page, we assume that links the user has clicked are more important to the user than links the user has not clicked. Researchers have used many methods to obtain the user's preference characteristics from the user's click behavior. For example, some researchers use an algorithm called Ranking SVM (Hersh et al. 1994) to obtain the best web page ranking for the user through the user's click information. In (Radlinski&Joachims 2005), the author not only extracts user preferences from a single user query, but also extracts user preferences from a series of user queries on the same information. These preferences are then trained through the improved algorithm of Ranking SVM . Sun et al. (2005) proposed an algorithm based on Singluar Value Decomposition, which improves the accuracy of the search engine's suggestion system by analyzing the user's click data. Attention time: Relatively speaking, attention time is a new type of implicit user feedback. Although it has been mentioned more and more in recent studies, it is still debated whether it can really reflect user intent. Kelly and Belkin (2004; 2001) suggest that there is not a very reliable correlation between the attention time of a document and its usefulness to users. But the difference is that in their research, attention time is measured by the average attention time of a group of users reading articles on different topics. Halabie et al. (2007) believed that for a user to pay attention to the time in the same search behavior, it can well reflect the user's preferences. We do not think the above two studies are contradictory, because the time of attention they calculate is not the same. In this paper, we assume that the attention time of a single user or a single topic can well reflect the user's preferences

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for sorting personalized webpages based on attention time.

The method for ranking webpage text and images based on user attention time includes the following steps:

1) Use the attention time to personalize the existing web page sorting, so that the sorting results conform to the user's psychology;

2) Use a custom browser to collect sample information of text attention time;

3) Use a custom browser to collect sample information on the attention time of pictures;

4) Calibrate the collected time-of-interest samples;

5) Predict the attention time of unknown web pages based on the similarity between text and pictures;

6) Using attention time combined with traditional search technology to generate personalized web pages and pictures for ranking.

The above-mentioned use of attention time to personalize the ranking of existing web pages, so that the ranking results conform to the user’s psychological steps: use the attention time as the source of user’s implicit feedback, so as to know the user’s preferences, and then analyze the user’s unbrowsed Predict the attention time of web pages or pictures, and finally sort the results according to the predicted attention time. The attention time is the reading or browsing time spent by users when browsing a web page or picture.

The steps of collecting the sample information of text attention time by using a custom browser: the client is a custom browser, for text search, on the search result page, the search engine usually sets A document provides a few lines of summary, tracking the mouse movement position, so as to record the time the user spends on a certain document, on the opened page, record the user's activity time on this page, and the attention time for this document is The time to read the summary is added to the time to read the entire document. If the user later returns to a page that has already been viewed, the attention time of the page will increase accordingly.

The steps of using a custom browser to collect the sample information of the picture attention time: the client is a custom browser, for picture search, the search engine will display the thumbnail of each picture on the result page, and the same Yes, the attention time is the time the user looks at the thumbnail plus the time the user looks at the original image. If a document has both text and pictures, its attention time is the sum of the two.

The step of correcting the collected time-of-interest samples: correcting the time-of-interest samples collected is as follows:

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; inf</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; raw</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; basic</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>$

Among them, t _att ^raw is the collected attention time, t _basic (u) is the time used by users to judge whether the document is worth reading, and t _att ^inf (u, d) is the potential attention time contained in the document d.

The time steps for predicting the attention time of unknown web pages based on the similarity between text and pictures:

a) Use Sim(d ₀ , d ₁ ) to represent the similarity between document d ₀ and document d ₁ , and Sim(d ₀ , d ₁ )[0, 1], before calculating the similarity between the two documents , delete advertisements, tags in the source code of the webpage, and the navigation bar on the webpage;

b) Express each training sample as {t _att (u, d _i )|i=1,...n}, where n is the number of documents that the current user has read, and the read documents are expressed as d _i (i=1,...,n), when the user encounters a new document d _x , calculate the similarity between the document d _x and all documents in the test set, and select k documents with the highest similarity , set k as min(10, n), the selected document is d _i (i=1,..., k), use the following equation to predict the attention time of d _x ,

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}$

It is used to control the proportion of the value of Sim(,), which is a small positive integer to prevent the denominator of the expression from being 0. The function (,) is used to remove some documents with very low similarity, which is defined for:

$\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; If</span>}{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0.01</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$

The steps of using attention time combined with traditional search technology to generate personalized web pages and pictures for sorting:

c) When a user submits a query request, the server first redirects the query to a traditional search engine and obtains the first n web pages returned. For each returned web page, the system will search for k web pages in the user's sample set that are related to The sample with the highest similarity of text or picture, and use the method in claim 8 to predict the attention time of the webpage;

d) For traditional sorting, the system will generate an attention time deviation, that is, in traditional sorting, documents with higher ranks will get higher attention time deviation. Use the following formula to define this deviation

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; attenuate</span>}}^{\mathrm{<span\; class="notranslate">\; offset</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Among them, rank(i) represents the ranking of document i in Google, and the parameter κ is used to control the slope of the attention time falling with the ranking;

e) Obtain the global attention time of document i from the attention time t _atten (i) and the deviation t _atten ^offset (i) of document i: $t_{\mathrm{attenuate}}^{\mathrm{overall}}\left(i\right)={\mathrm{\&kappa;}}_{\mathrm{overall}}{t}_{\mathrm{attenuate}}\left(i\right)+t_{\mathrm{attenuate}}^{\mathrm{offset}}\left(i\right),$ The parameter κ _overall is a user variable, which is used to control the proportion of the user's personalized ranking;

f) The final ranking will be in reverse order of total attention time.

The invention effectively combines the user's preferences in the search process, so that the final ranking result is closer to the ideal ranking expected by the user, so that the web page text and image search engine provides better personalized services for the user.

Description of drawings

Fig. 1 is the flowchart of specific embodiment;

Figure 2 is a screenshot of the custom browser in this example;

Figure 3 is the experimental results of 14 groups of text searches, and the specific data are in Table 2;

Figure 4 is the coordinate diagram of 7 groups of image search experiments, and the specific data are in Table 3 and Table 4; each group of experiments is conducted by different users under the same setting, and the coordinates are drawn on the coordinates of the user's average expected ranking , the smaller the average value, the higher the user-desired picture will appear in the search results.

Detailed ways

The method for ranking personalized web pages based on attention time includes the following steps:

1) Use the attention time to personalize the existing web page sorting, so that the sorting results conform to the user's psychology;

2) Use a custom browser to collect sample information of text attention time;

3) Use a custom browser to collect sample information on the attention time of pictures;

4) Calibrate the collected time-of-interest samples;

5) Predict the attention time of unknown web pages based on the similarity between text and pictures;

6) Using attention time combined with traditional search technology to generate personalized web pages and pictures for ranking.

The above-mentioned use of attention time to personalize the ranking of existing web pages, so that the ranking results conform to the user’s psychological steps: use the attention time as the source of user’s implicit feedback, so as to know the user’s preferences, and then analyze the user’s unbrowsed Predict the attention time of web pages or pictures, and finally sort the results according to the predicted attention time. The attention time is the reading or browsing time spent by users when browsing a web page or picture.

The steps of collecting the sample information of text attention time by using a custom browser: the client is a custom browser, for text search, on the search result page, the search engine usually sets A document provides a few lines of summary, tracking the mouse movement position, so as to record the time the user spends on a certain document, on the opened page, record the user's activity time on this page, and the attention time for this document is The time to read the summary is added to the time to read the entire document. If the user later returns to a page that has already been viewed, the attention time of the page will increase accordingly.

The steps of using a custom browser to collect the sample information of the picture attention time: the client is a custom browser, for picture search, the search engine will display the thumbnail of each picture on the result page, and the same Yes, the attention time is the time the user looks at the thumbnail plus the time the user looks at the original image. If a document has both text and pictures, its attention time is the sum of the two.

The step of correcting the collected time-of-interest samples: correcting the time-of-interest samples collected is as follows:

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; inf</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; raw</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; basic</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>$

Among them, t _att ^raw is the collected attention time, t _basic (u) is the time used by users to judge whether the document is worth reading, and t _att ^inf (u, d) is the potential attention time contained in the document d.

The time steps for predicting the attention time of unknown web pages based on the similarity between text and pictures:

a) Use Sim(d ₀ , d ₁ ) to represent the similarity between document d ₀ and document d ₁ , and Sim(d ₀ , d ₁ )[0, 1], before calculating the similarity between the two documents , delete advertisements, tags in the source code of the webpage, and the navigation bar on the webpage;

b) Express each training sample as {t _att (u, d _i )|i=1,...n}, where n is the number of documents that the current user has read, and the read documents are expressed as d _i (i=1,...,n), when the user encounters a new document d _x , calculate the similarity between the document d _x and all documents in the test set, and select k documents with the highest similarity , set k as min(10, n), the selected document is d _i (i=1,..., k), use the following equation to predict the attention time of d _x ,

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}$

It is used to control the proportion of the value of Sim(,), which is a small positive integer used to prevent the denominator of the expression from being 0. The function (,) is used to remove some documents with very low similarity. It is defined for:

$\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; If</span>}{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0.01</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$

The steps of using attention time combined with traditional search technology to generate personalized web pages and pictures for sorting:

c) When a user submits a query request, the server first redirects the query to a traditional search engine and obtains the first n web pages returned. For each returned web page, the system will search for k web pages in the user's sample set that are related to The sample with the highest similarity of text or picture, and use the method in claim 8 to predict the attention time of the webpage;

d) For traditional sorting, the system will generate an attention time deviation, that is, in traditional sorting, documents with higher ranks will get higher attention time deviation. Use the following formula to define this deviation

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; attenuate</span>}}^{\mathrm{<span\; class="notranslate">\; offset</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Among them, rank(i) represents the ranking of document i in Google, and the parameter κ is used to control the slope of the attention time falling with the ranking;

e) Obtain the global attention time of document i from the attention time t _atten (i) and the deviation t _atten ^offset (i) of document i: $t_{\mathrm{attenuate}}^{\mathrm{overall}}\left(i\right)={\mathrm{\&kappa;}}_{\mathrm{overall}}{t}_{\mathrm{attenuate}}\left(i\right)+t_{\mathrm{attenuate}}^{\mathrm{offset}}\left(i\right),$ The parameter κ _overall is a user variable, which is used to control the proportion of the user's personalized ranking;

f) The final ranking will be in reverse order of total attention time.

Example:

The flow structure of the web page text and image sorting method based on the user's attention time of the present invention is shown in FIG. 1 . The personalized sorting system includes two parts, the client and the server, the client 20, a custom browser to obtain the user's attention time, the server includes 30, a sample collection module, 40, attention time correction, 50, user database and 60 . Document database, 70. query interface, 80. traditional engine module, 90. document preprocessing module, 100. document comparison module, 110. attention time prediction module, 120. sorting module. .

Customize the browser 20 to track and analyze the user's mouse movement, and finally obtain the user's attention time on each document. In this example, pair attention times recorded by a custom browser developed by us are presented (Fig. 2).

The sample collection module 30 stores the sample data sent by the client into the database corresponding to the user, and if a certain document does not exist in the document database, downloads and stores it into the document database.

The attention time correction module 40 needs to correct the predicted attention time obtained directly from the client. When a user browses a document, no matter whether the document is useful to the user or not, the user has to spend a period of time to roughly browse the document. Generally speaking, the attention time obtained at this time includes both the user's actual attention time and the user's rough browsing time of the document. In order to overcome this problem, we use the following equation to correct our original attention time:

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; inf</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; raw</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; basic</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>$

t _att ^raw is the attention time we obtained originally, t _basic (u) is the time used by the user to judge whether the document is worth reading, and t _att ^inf (u, d) is the potential attention time contained in the document d.

The user database 50 stores the attention time of each user of the system on the document, which is stored in MYSQL in this example.

The document database 60 stores the data of documents (text web pages and pictures), and uses MYSQL to store in this example.

The query interface 70 provides a web portal for user query, and provides two services of text search and image search. In this example, the query interface is implemented with jsp.

Traditional engine module 80, when a user submits a query request, the server will analyze the result page of a traditional search engine (such as Google) and obtain the first 300 documents in the returned results, and download and save the documents to the document server.

In the document preprocessing module 90, the web pages downloaded directly from the website contain a lot of useless information, such as HTML tags, advertisement columns, navigation bars and so on. This module is used to remove useless information in web pages and keep the main documents that users will pay attention to. In this example, we implemented the function of removing HTML tags.

In the document comparison module 100, the text similarity algorithm selected is the extended Jaccard method (Tanimoto); the image similarity algorithm selected is the similarity algorithm based on "Auto Color Correlogram" (Huang et al.1997).

Pay attention to the time prediction module 110, including the following steps:

a.) This module predicts the attention time for each document in the traditional engine module. First, we denote each training sample as {t _att (u, d _i )|i=1,...n}, where n is the number of documents that the current user has read. Read documents are denoted as d _i (i=1, . . . , n). For the document d _x returned by the traditional engine, we will calculate the similarity between the document d _x and all documents in the test set. We then pick the k documents with the highest similarity. In our experiments, we set k as min(10,n). The documents we selected are d _i (i=1,...,k). Then we use the following equation to predict the attention time of d _x .

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}$

Among them, γ is used to control the proportion of the value of Sim(,), and ε is a small positive integer to prevent the denominator of the expression from being 0. The function δ(,) is used to remove some documents with very low similarity, which is defined as

$\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; If</span>}{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0.01</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.$

b.) In the initial stage of system operation, we will also convert the ranking of traditional engines into a focus on time deviation. We convert traditional rankings into a normalized attention-time bias with values between 0 and 1 using the following equation:

${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; attenuate</span>}}^{\mathrm{<span\; class="notranslate">\; offset</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}$

Where rank(i) represents the ranking of document i in traditional search engines. The reason why we choose such a formula is that it can convert the page ranking information into attention time, and the attention time obtained by conversion of lower ranking documents is relatively shorter. The parameter κ _d is used to control the slope of the attention time decreasing with the ranking. In our experiment, we set it to 0.2.

c.) Once we have the attention time t _atten (i) and the bias t _atten ^offset i for document i, we can obtain the global attention time for that document: $t_{\mathrm{attenuate}}^{\mathrm{overall}}i={\mathrm{\&kappa;}}_{\mathrm{overall}}{t}_{\mathrm{attenuate}}i+t_{\mathrm{attenuate}}^{\mathrm{offset}}i.$ The parameter κ _overall is a user variable, which is used to control the proportion of the user's personalized ranking. The ranking of the final web pages is arranged in descending order of global attention time. We implemented a method to automatically set the value of κ _overall . When there are few samples in the training set, the value of κ _overall is small, and when the number of samples in the training set becomes more and more, the value of κ _overall becomes larger and larger. . The reason for this is that our sorting algorithm is fundamentally a learning algorithm. However, like other learning algorithms, when the training sample set is still small, the algorithm will produce poor results, so we need to learn from the ranking results of traditional engines. In our experiments, we used a sigmoid function to automatically verify the value of κ _overall and found it to be a constant, usually 0.1.

The sorting module 120, the sorting module sorts the results in reverse order according to the global attention time of all documents, and returns the results to the user.

The experimental results of Tables 1 to 4 clearly show the superiority of this method;

Table 1 shows the respective rankings of the top 17 texts obtained by using "Web search technology" (Web search technology) as the keyword text search; each column is the user's ideal ranking from left to right, and the web page search engine Google ranking, and the ranking after the user has read 2, 5, 8, 10, and 15 web pages; the last row indicates the sum of the absolute errors between each ranking and the user's ideal ranking;

Table 1

Rk _user Rk _Google Rk ₂ Rk ₅ Rk ₈ Rk ₁₀ Rk ₁₅ 6 1 1 15 13 11 7 9 4 17 16 14 12 9 1 2 2 1 1 1 1 17 3 10 17 16 16 16 2 6 3 7 2 2 2 15 5 12 9 15 15 15 16 7 13 14 17 17 17 5 8 9 4 12 10 6 11 15 13 6 6 14 11 10 14 15 13 11 13 10 14 18 16 12 9 7 14 12 16 12 11 10 9 12 3 4 4 4 3 3 3 13 11 9 10 8 8 13 8 6 5 3 4 4 8 7 5 14 8 7 6 5 4 7 8 5 5 5 4 0 96 60 52 44 42 6

Table 2 is the experimental data of 14 different users doing text searches on different keywords; each row represents the sum of the absolute errors in the rankings between the rankings obtained in each group of experiments and the user's ideal rankings, and these data are also displayed graphically in Figure 3;

Table 2

query word Rk _Google Rk ₂ Rk ₅ Rk ₈ Rk ₁₀ Rk ₁₅ Greenhouse effect 88 66 66 62 52 16 Keno Linux operating system (Gnome Linux) 86 64 60 56 50 18 encryption algorithm 123 99 78 65 45 twenty two Reduced Instruction Set (RISC) 94 82 62 58 50 32 Advertising ethics 94 77 62 47 41 10 Da Vinci (da Vinci) 103 99 65 49 39 twenty one olympic 77 72 58 50 36 10 Anckor 90 94 92 74 46 16 Color management 128 146 94 90 66 52 National Basketball Association (NBA) 109 94 76 62 36 twenty two Correlation 122 114 114 88 82 54 Houston 133 98 92 85 76 43 investment 132 120 104 100 94 46 Samsung (samsung) 71 74 68 42 36 4

Table 3 is a group of image search experiment data with "Picasso" (Picasso) as the keyword; the user wants to use "Picasso" to find Picasso's self-portrait, and only 6 of the 60 pictures meet the user's needs; One column indicates the rankings of the images that meet the requirements in the image search engine Google and this method; Rk1st, Rk2nd, and Rk3rd respectively indicate the rankings of the users after reading pages 1, 2, and 3 of the search results situation; Rkgoogle represents the ranking of these images in Google Image Search. The last line is the average ranking of these images in each case; the smaller the average ranking value, the earlier the image required by the user will appear;

table 3

Rk _Google Rk _1st Rk _2nd Rk _3rd 9 1 1 1 16 63 3 5 17 3 2 3 twenty three 41 15 2 41 twenty four 37 4 48 13 4 6 25.67 24.17 10.33 3.5

Table 4 is the experimental data of another 6 groups of image search; in the experiment, each user is required to find the image he needs in 60 pictures; the first column is the search keyword, and the second column is the number of pictures that the user needs. RkGoogle indicates the average ranking of the web pages required by the user in the web image search engine Google; Rk1st, Rk2nd, and Rk3rd respectively indicate the average web page images required by the user after reading the first, second, and third pages of the search results ranking;

Table 4

query word #Images Rk _Google Rk _1st Rk _2nd Rk _3rd Tree 9 16.22 10.56 8.11 6 Desert 10 20.2 15.4 14.1 12.6 South Pole 14 24.57 23.5 21.21 13.71 Apple 9 21.33 12.33 11.78 11.22 break heart 5 22.4 twenty two 21.2 8.2 Pirates of the Caribbean 19 24.37 27.16 20.37 15.32

The above table shows that the present invention effectively combines the user's preferences in the search process, making the final ranking result closer to the ideal ranking expected by the user, so that the web text and image search engine can provide users with better personalized services.

The above descriptions are only preferred embodiments of the attention time-based user-oriented personalized web page sorting method and system of the present invention, and are not intended to limit the scope of the substantive technical content of the present invention. The user-oriented personalized web page sorting method and system based on attention time of the present invention, its essential technical content is broadly defined in the claims, any technical entity or method completed by others, if it is the same as defined in the claims or identical or equivalent changes will be deemed to be covered within the scope of this patent protection.

Claims (7)

1. A webpage text and image sorting method based on user attention time, it is characterized in that comprising the following steps: 1) Use the attention time to personalize the existing web page sorting, so that the sorting results conform to the user's psychology; 2) Use a custom browser to collect sample information of text attention time; 3) Use a custom browser to collect sample information on the attention time of pictures; 4) Calibrate the collected time-of-interest samples; 5) Predict the attention time of unknown web pages based on the similarity between text and pictures; 6) Using attention time combined with traditional search technology to generate personalized web pages and pictures for ranking. 2. a kind of web page text and image sorting method based on user attention time according to claim 1, it is characterized in that described utilization time of attention is carried out individualized improvement to existing web page sorting, makes sorting result accord with user's psychological step: The attention time is used as the source of the user's implicit feedback, so as to know the user's preferences, and then predict the attention time of the web pages or pictures that the user has not browsed, and finally sort the results according to the predicted attention time. The attention time is the user's The reading or browsing time spent viewing a web page or image. 3. a kind of webpage text and image sorting method based on user's attention time according to claim 1, it is characterized in that described utilizes self-defined browser, collects the sample information step of text attention time: client is an automatic Defined browsers, for text search, on the search results page, the search engine usually provides a few lines of summary for each document on the search results page, tracking the mouse movement position, so as to record the user's spending on a document Time, on the opened page, record the user's activity time on this page. The attention time for this document is the time of reading the summary plus the time of reading the entire document. If the user returns to the page that has been read later , then the page's attention time will increase accordingly. 4. a kind of webpage text and image sorting method based on user's attention time according to claim 1, it is characterized in that described utilize self-definition browser, collect the sample information step of picture attention time: client is an automatic Defined browsers, for image search, the search engine will display the thumbnail of each image on the result page. Similarly, the attention time is the time the user looks at the thumbnail plus the time the user looks at the original image. If a document is both There are text and pictures, and its attention time is the sum of the two. 5. A kind of webpage text and image ordering method based on user attention time according to claim 1, it is characterized in that described step of correcting the attention time sample collected: the correction formula is as follows to the attention time sample collected: ${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; inf</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}^{\mathrm{<span\; class="notranslate">\; raw</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; basic</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>$ Where t _att ^raw is the collected attention time, t _basic (u) is the time used by users to judge whether the document is worth reading, and t _att ^inf (u, d) is the potential attention time contained in the document d. 6. a kind of webpage text and image sorting method based on user attention time according to claim 1, it is characterized in that described based on text and picture similarity to predict the attention time step of unknown webpage: a) Use Sim(d ₀ , d ₁ ) to represent the similarity between document d ₀ and document d ₁ , and Sim(d ₀ , d ₁ )[0, 1], before calculating the similarity between the two documents , delete advertisements, tags in the source code of the webpage, and the navigation bar on the webpage; b) Express each training sample as {t _att (u, d _i )|i=1,...n}, where n is the number of documents that the current user has read, and the read documents are expressed as d _i (i=1,...,n), when the user encounters a new document d _x , calculate the similarity between the document d _x and all documents in the test set, and select k documents with the highest similarity , set k as min(10, n), the selected document is d _i (i=1,..., k), use the following equation to predict the attention time of d _x , ${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; att</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}$ It is used to control the proportion of the value of Sim(,), which is a small positive integer to prevent the denominator of the expression from being 0. The function (,) is used to remove some documents with very low similarity, which is defined for: $\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; If</span>}{\mathrm{<span\; class="notranslate">\; Sim</span>}}^{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0.01</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$ 7. a kind of web page text and image sorting method based on user attention time according to claim 1, it is characterized in that described utilizing attention time in conjunction with traditional search technology to generate personalized webpage and picture sorting step: c) When a user submits a query request, the server first redirects the query to a traditional search engine and obtains the first n web pages returned. For each returned web page, the system will search for k web pages in the user's sample set that are related to The sample with the highest similarity of text or picture, and use the method in claim 8 to predict the attention time of the webpage; d) For traditional sorting, the system will generate an attention time deviation, that is, in traditional sorting, documents with higher rankings will get higher attention time deviation. Use the following formula to define this deviation ${\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; attenuate</span>}}^{\mathrm{<span\; class="notranslate">\; offset</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&kappa;</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; rank</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$ Among them, rank(i) represents the ranking of document i in Google, and the parameter κ is used to control the slope of the attention time decreasing with the ranking; e) Obtain the global attention time of document i from the attention time t _atten (i) and the deviation t _atten ^offset (i) of document i: $t_{\mathrm{attenuate}}^{\mathrm{overall}}\left(i\right)={\mathrm{\&kappa;}}_{\mathrm{overall}}{t}_{\mathrm{attenuate}}\left(i\right)+t_{\mathrm{attenuate}}^{\mathrm{offset}}\left(i\right),$ The parameter κ _overall is a user variable, which is used to control the proportion of the user's personalized ranking; f) The final ranking will be in reverse order of total attention time.

Images