CN1845100A - Image extraction feedback method in web search - Google Patents

Abstract

The invention discloses an image summary feedback method used in Web search, which comprises: (1) the page processor receives the inquiry formed by the keywords referred by user, using Web search method based on the keyword to search the Web; (2), using the method that based on image summary to feedback, to extract the user inquiry; (3), feeding back demanded search result. The invention can utilize the word information and the image information in the website, to effectively attain the information of user to improve the ability of Web search.

Description

Image extraction feedback method in the Web search

One, technical field

The present invention relates to page processor, particularly a kind of image extraction feedback method that is applied to the Web search.

Two, background technology

Along with developing rapidly of multimedia technology and internet, Web becomes people most important and information source and the most frequently used information exchange platform efficiently gradually.Because the quantity of information of Web itself is huge and content is complicated, user's searching information from the internet is very difficult, and therefore, designing effective Web search technique becomes an important job.In the Web search procedure, after search engine feeds back to the user with Search Results, if the user is dissatisfied to Search Results, usually will provide feedback according to Search Results, it is relevant with ferret out for example to indicate which result, search engine utilizes these feedback informations further to search for again, to produce better Search Results.User feedback technology in the present Web search has only been used the Word message in the webpage, and does not utilize a large amount of image informations that exist in the webpage.

Three, summary of the invention

1, goal of the invention: fundamental purpose of the present invention is that the user feedback technology in searching at present Web is not utilized this problem of picture material in the webpage well, provide a kind of user feedback method of utilizing Word message and image information simultaneously, by the performance of the auxiliary Web of the raising search of the information requirement of obtaining the user effectively.

2, technical scheme: for realizing purpose of the present invention, a kind of image extraction feedback method that is applicable in the Web search of the present invention, may further comprise the steps: the inquiry of forming by keyword that (1) page processor accepts that the user submits to, use is searched for Web based on the Web searching method of keyword; (2) method of feeding back based on the image extraction user inquiring of refining; (3) finally return customer satisfaction system Search Results.

Wherein the method for feeding back based on the image extraction user inquiring of refining may further comprise the steps: (11) obtain user's searching keyword; (12) utilize Web searching method to obtain relevant webpage based on keyword; (13) literal that generates webpage from the related web page that searches is made a summary and image extraction, and these information are submitted to the user, understands Search Results for the user; (14) user judges the webpage that searches, if webpage satisfies user's demand, then goes to (16), and whole process finishes; Otherwise, go to (15); (15) obtain user's feedback information, generate new searching keyword, go to (12); (16) finish.

3, beneficial effect: method provided by the present invention is used a large amount of image informations that exist in Word message in the webpage and the webpage simultaneously, the performance of the greatly auxiliary Web of the raising search of information requirement by obtaining the user effectively.

Below in conjunction with accompanying drawing most preferred embodiment is elaborated.

Four, description of drawings

Fig. 1 is the page processor workflow diagram.

Fig. 2 is the process flow diagram of the inventive method.

Fig. 3 is the process flow diagram that obtains literal summary and image extraction.

Fig. 4 is the process flow diagram that obtains first kind image extraction.

Fig. 5 is the process flow diagram that obtains the second class image extraction.

Fig. 6 obtains the process flow diagram that field feedback generates new searching keyword.

Fig. 7 is the process flow diagram that the user selects to carry out first kind feedback.

Fig. 8 is the process flow diagram that the user selects to carry out second class feedback.

Five, embodiment

As shown in Figure 1, the inquiry of being made up of keyword that page processor accepts that the user submits to uses the Web search technique based on keyword of classics that Web is searched for.Use the technology of describing among Fig. 2 of feeding back based on the image extraction user inquiring of refining, finally return customer satisfaction system Search Results.

Technology of the present invention as shown in Figure 2.Step 10 is initial actuatings.Step 11 is obtained user's searching keyword.Step 12 utilizes the classical Web search technique based on keyword to obtain relevant webpage.Step 13 generates the literal summary and the image extraction of webpage from the related web page that searches, and these information are submitted to the user, understands Search Results for the user.The literal summary is meant the content of text in the webpage is concentrated the resulting literal that can reflect web page contents to a certain extent in back that classical Web search technique utilizes the literal summary to come to the user feedback Search Results exactly.Image extraction is the technology that this patent proposes, specifically be meant from webpage, obtain with the maximally related image of user inquiring, these image feedback will be helped the user to the user is faster to understand Search Results more accurately.Step 14 user judges the webpage that searches, if webpage satisfies user's demand, then goes to step 16, and whole process finishes; Otherwise, go to step 15, obtain user's feedback information, utilize these information to generate new searching keyword, and new searching keyword is submitted to system, go to step 12, repeat above process.

Fig. 3 describes step 13 in detail, promptly obtains the text snippet and the image extraction of webpage.Step 130 is initial steps.Step 131 uses the method for classical generation literal summary to generate the literal summary of webpage.Step 132 generates the image extraction of the first kind, and the image extraction of this class is corresponding one by one with webpage, is used for auxiliary literal summary displayed web page content and is used for user feedback afterwards.Step 133 generates the image extraction of second class, the feedback of the user after such image extraction is used for.In the Web search interface, first kind image extraction should be submitted to the user together with the literal summary of corresponding webpage, and the second class image extraction does not then need to put together with the literal summary.Step 134 finishes.

Fig. 4 describes step 132 in detail, promptly obtains first kind image extraction.Step 1320 is initial steps.Step 1321 is changed to 1 with counter i.Step 1322 is judged the number M of the i webpage whether great-than search arrives.If i greater than M, goes to step 1328, finish; Otherwise go to step 1323.Step 1323 utilizes classical webpage block algorithm that i webpage carried out piecemeal.Step 1324 utilizes the importance method of discrimination in the classical Web search technique to judge that whether important each piecemeal is.Step 1325 is selected the piecemeal the most similar to searching keyword from the important piecemeal that comprises piece image at least.Here adopt classical literal method for measuring similarity to measure similarity between the text description of searching keyword and piecemeal.Step 1326 is selected the image extraction of the image of appearance at first as i webpage from the most similar piecemeal.Step 1327 adds 1 with i, goes to step 1322.

Fig. 5 describes step 133 in detail, promptly obtains the second class image extraction.Step 1330 is initial steps.Step 1331 will be gathered IMG and put sky.Step 1332 is changed to 1 with counter i.Step 1333 is judged the number M of the i webpage whether great-than search arrives.If i greater than M, goes to step 1338; Otherwise go to step 1334.Step 1334 pair utilizes the webpage method of partition in the classical Web search technique that i webpage carried out piecemeal.Step 1335 utilizes the importance method of discrimination in the classical Web search technique to judge that whether important each piecemeal is.Step 1336 adds set IMG with the image in the important piecemeal.Step 1337 adds 1 with i, goes to step 1333.Step 1338 sorts to the image among the IMG according to the text description of image among the IMG and the similarity degree of searching keyword, selects the image extraction of the most similar j width of cloth image as second class.The measuring similarity here adopts classical literal method for measuring similarity.The text description of image is made up of three parts, the ALT field of image in html source code, the text description of the title of image place webpage and image place piecemeal.Step 1339 finishes.

Fig. 6 describes step 15 in detail, promptly obtains user's feedback information, generates new searching keyword with the inquiry of refining.Step 150 is initial steps.Which kind of feedback is step 151 selected to carry out by the user.Step 152, the user selects to carry out first kind feedback, according to user's the new searching keyword of feedback information generation.First kind feedback is based on that the image extraction of the first kind finishes.In the feedback of this class, the user judges the webpage phase WICCON whether in cus toms clearance or not that searches in conjunction with the literal summary and the image extraction that provide.Step 153, the user selects to carry out second class feedback, according to user's the new searching keyword of feedback information generation.Second class feedback is based on that the image extraction of second class finishes.In the feedback of this class, the user judges directly whether the image extraction that provides is relevant.Step 154 finishes.

Fig. 7 describes step 152 in detail, promptly generates new searching keyword in first kind feedback.Step 1520 is initial steps.Step 1521 is put set TERM for empty.It is 1 that step 1522 is put counter i.Step 1523 judges that i is whether less than the number F of the webpage of user feedback.If i greater than F, goes to step 1526; Otherwise go to step 1524.Step 1524 adds set TERM (appeared in the searching keyword except) with the speech that occurs in i the webpage.Step 1525 adds 1 with counter i, goes to step 1523.Step 1526 according to the score value of each speech among formula (1) the calculating TERM, selects k the highest speech of score value as newly-increased searching keyword.In the formula (1), Score (t) is the score value of speech t, r _tBe chosen as the number that comprises t in the relevant webpage, n by the user _tBe the number that comprises t in the webpage that searches, R is chosen as the number of relevant webpage by the user, and N is the number of the webpage that searches.Step 1527 finishes.

$\mathrm{Score}\left(t\right)=\log \frac{r_t/(R-r_t)}{(n_t-r_t)/(N-n_t-R+r_t)}\×(\frac{r_t}{R}-\frac{n_t-r_t}{N-R})---\left(1\right)$

Fig. 8 describes step 153 in detail, promptly generates new searching keyword in second class feedback.Step 1530 is initial steps.Step 1531 is put set TERM for empty.It is 1 that step 1532 is put counter i.Step 1533 judges that i is whether less than the number F of the image of user feedback.If i greater than F, goes to step 1536; Otherwise go to step 1534.Step 1534, the speech that occurs in the textual description with i image add set TERM (appeared in the searching keyword except).Step 1535 adds 1 with counter i, goes to step 1533.Step 1536 according to the score value of each speech among formula (1) the calculating TERM, selects k the highest speech of score value as newly-increased searching keyword.Note in formula this moment (1) r _tBe chosen as the number that comprises the image of t in the textual description of relevant image by the user, n _tBe that Fig. 5 gathers the number that comprises the image of t in the textual description of the image among the IMG, R is chosen as the number of relevant image by the user, and N is the number that Fig. 5 gathers the image among the IMG.

Claims (7)

1, the image extraction feedback method in a kind of Web search is characterized in that this method may further comprise the steps:

(1) the page processor inquiry of forming by keyword of accepting that the user submits to, use is searched for Web based on the Web searching method of keyword;

(2) method of feeding back based on the image extraction user inquiring of refining;

(3) finally return customer satisfaction system Search Results.

Wherein the method for feeding back based on the image extraction user inquiring of refining may further comprise the steps:

(11) obtain user's searching keyword;

(12) utilize Web searching method to obtain relevant webpage based on keyword;

(13) literal that generates webpage from the related web page that searches is made a summary and image extraction, and these information are submitted to the user, understands Search Results for the user;

(14) user judges the webpage that searches, if webpage satisfies user's demand, then goes to (16), and whole process finishes; Otherwise, go to (15);

(15) obtain user's feedback information, generate new searching keyword, go to (12);

(16) finish.

2, the image extraction feedback method in the Web search according to claim 1 is characterized in that the text snippet and the image extraction that obtain webpage in (13) may further comprise the steps:

(131) use the literal of the method generation webpage of known generation literal summary to make a summary;

(132) image extraction of the generation first kind, the image extraction of this class is corresponding one by one with webpage, is used for auxiliary literal summary displayed web page content and is used for user feedback afterwards;

(133) image extraction of generation second class, the feedback of the user after such image extraction is used for;

(134) finish.

3, the image extraction feedback method in the Web search according to claim 2 is characterized in that obtaining in (132) first kind image extraction and may further comprise the steps:

(1321) counter i is changed to 1;

(1322) judge the number M of the i webpage whether great-than search arrives,, otherwise go to (1323) if i, goes to (1328) greater than M;

(1323) utilize known webpage block algorithm that i webpage carried out piecemeal;

(1324) utilize the importance method of discrimination in the Web searching method to judge that whether important each piecemeal is;

(1325) from the important piecemeal that comprises piece image at least, select the piecemeal the most similar to searching keyword;

(1326) from the most similar piecemeal, select the image extraction of the image of appearance at first as i webpage;

(1327) i is added 1, go to step 1322;

(1328) finish.

4, the image extraction feedback method in the Web search according to claim 2 is characterized in that the step of obtaining the second class image extraction in (133) is:

(1331) will gather IMG and put sky;

(1332) counter i is changed to 1;

(1333) judge the number M of the i webpage whether great-than search arrives, if i, goes to (1338) greater than M; Otherwise go to (1334);

(1334) utilize the webpage method of partition in the Web searching method that i webpage carried out piecemeal;

(1335) utilize the importance method of discrimination in the Web searching method to judge that whether important each piecemeal is;

(1336) image in the important piecemeal is added set IMG;

(1337) i is added 1, go to (1333);

(1338) according to the text description of image among the IMG and the similarity degree of searching keyword the image among the IMG is sorted, select the image extraction of the most similar j width of cloth image as second class;

(1339) finish.

5, the image extraction feedback method in the Web according to claim 1 search is characterized in that obtaining in (15) user's feedback information, generates new searching keyword and inquires about to refine and may further comprise the steps;

(151) select to carry out which kind of feedback by the user;

(152) user selects to carry out first kind feedback, according to user's the new searching keyword of feedback information generation;

(153) user selects to carry out second class feedback;

(154) finish.

6, the image extraction feedback method in the Web search according to claim 5 is characterized in that generating new searching keyword in (152) in first kind feedback may further comprise the steps:

(1521) put set TERM for empty;

(1522) putting counter i is 1;

(1523) judge that i is whether less than the number F of the webpage of user feedback, if i, goes to (1526) greater than F; Otherwise go to (1524);

(1524) speech that occurs in i the webpage is added set TERM, appeared in the searching keyword except;

(1525) counter i is added 1, go to (1523);

(1526), select k the highest newly-increased searching keyword of speech conduct of score value according to the score value of each speech among the following formula calculating TERM:

$\mathrm{Score}\left(t\right)=\log \frac{r_t/(R-r_t)}{(n_t-r_t)/(N-n_t-R+r_t)}\×(\frac{r_t}{R}-\frac{n_t-r_t}{N-R})$

In the formula, Score (t) is the score value of speech t, r _tBe chosen as the number that comprises t in the relevant webpage, n by the user _tBe the number that comprises t in the webpage that searches, R is chosen as the number of relevant webpage by the user, and N is the number of the webpage that searches;

(1527) finish.

7, the image extraction feedback method in the Web search according to claim 5 is characterized in that generating new searching keyword in (153) in second class feedback may further comprise the steps:

(1531) put set TERM for empty;

(1532) putting counter i is 1;

(1533) judge that i is whether less than the number F of the image of user feedback, if i, goes to (1536) greater than F; Otherwise go to (1534);

(1534) speech that occurs in the textual description with i image adds set TERM, appeared in the searching keyword except;

(1535) counter i is added 1, go to (1533);

(1536), select k the highest newly-increased searching keyword of speech conduct of score value according to the score value of each speech among the following formula calculating TERM:

$\mathrm{Score}\left(t\right)=\log \frac{r_t/(R-r_t)}{(n_t-r_t)/(N-n_t-R+r_t)}\×(\frac{r_t}{R}-\frac{n_t-r_t}{N-R})$

(1537) finish.

Images