CN102609458B - A kind of picture recommendation method and device - Google Patents
A kind of picture recommendation method and device Download PDFInfo
- Publication number
- CN102609458B CN102609458B CN201210009043.7A CN201210009043A CN102609458B CN 102609458 B CN102609458 B CN 102609458B CN 201210009043 A CN201210009043 A CN 201210009043A CN 102609458 B CN102609458 B CN 102609458B
- Authority
- CN
- China
- Prior art keywords
- picture
- keyword
- word
- text
- target photo
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application provides a kind of picture recommendation method and device, method wherein specifically comprises: the inquiry request receiving user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo; According to the web page text at described result picture place, extract the keyword describing picture semantic feature, as the keyword of this Target Photo; In search daily record, carry out the coupling of keyword, and the respective objects picture with Keywords matching is recommended user; The Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword.The application can provide the picture agreeing with users ' individualized requirement, the acquisition channel of extending user information of interest.
Description
Technical field
The application relates to picture Processing Technique field, particularly relates to a kind of picture recommendation method and device.
Background technology
At present along with the development of network technology, user has no longer been satisfied with just to the search of text the requirement of search engine, and a lot of user also wishes to be searched for network picture by search engine.
Current photographic search engine mostly adopts text based search technique, and this technology, using the object of picture as database purchase, is described with key word.But for the visual signature comprised in picture, as color or shape etc., cannot be described with text, like this, when needing according to the visual feature search picture comprised in picture, text based search technique will be no longer applicable.Such as, user often runs into such problem, website or computer are seen a picture comprising article, but and do not know what the article in this picture are, therefore be difficult to the visual signature of these article to describe in words out, even if its visual signature has described out by the good user of ability to express, be also difficult to find the picture with this picture analogies in existing search engine, cause search efficiency low, use network traffics larger.
The problem low for above-mentioned search efficiency, use network traffics are larger, some photographic search engines provide to scheme to search figure function, to scheme searching figure function, picture consistent for vision content should be returned to user, to meet some search need of user.Such as certain user likes collection picture, and least patient is exactly have watermark above beautiful figure, as long as uploading pictures is to photographic search engine, and the picture just can found not with watermark of clicking; And for example, little picture can be uploaded, search out each version of this little picture, as clear large figure etc.
Also have some photographic search engines while scheming searching figure function, to provide picture recommendation function providing, with reference to Fig. 1, show the process flow diagram of picture recommendation method in a kind of photographic search engine of prior art, specifically can comprise:
Step 101, submit queries picture;
The visual signature such as color, texture, shape of step 102, extraction inquiry picture;
Step 103, the visual signature of picture in the inquiry visual signature of picture and database is carried out similarity comparison;
Step 104, vision similar pictures is recommended user.
Because the visual signature comparison such as color, texture, shape of picture recommendation results foundation obtains, therefore the similar main finger outward appearance of vision is here similar, such as user uploads girl can the graceful picture in Jede, in picture, girl can the graceful hair color in Jede be golden, then photographic search engine may return the similar picture containing golden tresses of vision, as the picture of blondie, the picture of Cibotium barometz (L.) J. Sm sometimes even can be returned, etc.
But some user exists some individual demands, as user uploads the picture of Liu Dehua, the picture such as film poster, individual's description seeing Liu Dehua also may be wished.Now, the Search Results that in prior art, vision content the is consistent picture recommendation results similar with vision all can not meet the individual demand of user.
In a word, the technical matters needing those skilled in the art urgently to solve is exactly: how can provide the picture agreeing with users ' individualized requirement.
Summary of the invention
Technical problems to be solved in this application are to provide a kind of picture recommendation method and device, can provide the picture agreeing with users ' individualized requirement, the acquisition channel of extending user information of interest.
In order to solve the problem, this application discloses a kind of picture recommendation method, comprising:
Receive the inquiry request of user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
According to the web page text at described result picture place, extract the keyword describing picture semantic feature, as the keyword of this Target Photo;
In search daily record, carry out the coupling of keyword, and the respective objects picture with Keywords matching is recommended user; The Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword.
Preferably, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, comprising:
According to the result of described web page text being carried out to cluster analysis, remove web page text isolated in described web page text, obtain remaining text;
Extract word frequency in described residue text the highest and there is word or the phrase of practical significance, as the keyword describing picture semantic feature.
Preferably, word or the phrase in described residue text with practical significance is extracted by following steps:
Call the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retain institute's predicate or phrase; Described entity word stock contains the entity word with practical significance.
Preferably, word or the phrase in described residue text with practical significance is extracted by following steps:
Extract word or the phrase in described residue text with practical significance according to part of speech, described extraction process comprises:
When word in described residue text or phrase are any one in interjection, pronoun or auxiliary words of mood, abandon institute's predicate or phrase.
Preferably, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, also comprises:
According to the adjacent co-occurrence frequency of other vocabulary in described keyword and described residue text, add up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
Preferably, described method also comprises:
At the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtain remaining picture;
The described step respective objects picture with Keywords matching being recommended user is that described residue picture is recommended user.
Preferably, the described step respective objects picture with Keywords matching being recommended user, comprising:
According to described search daily record, the on-line query request number that statistics is described corresponding with the respective objects picture of Keywords matching;
According to the descending of on-line query request number, the respective objects picture with Keywords matching is recommended user.
Preferably, this Target Photo is the picture that the query strategy corresponding with this inquiry request mates most; Described result picture is other pictures being greater than matching threshold except Target Photo.
On the other hand, disclosed herein as well is a kind of picture recommendation apparatus, comprising:
Picture searching module, for receiving the inquiry request of user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
Keyword abstraction module, for the web page text according to described result picture place, extracts the keyword describing picture semantic feature, as the keyword of this Target Photo;
Matching module, for carrying out the coupling of keyword in search daily record; The Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword; And
Picture recommending module, for recommending user by the respective objects picture with Keywords matching.
Preferably, described keyword abstraction module comprises:
Remove submodule, for according to the result of described web page text being carried out to cluster analysis, remove web page text isolated in described web page text, obtain remaining text; And
Extract submodule, the highest and there is word or the phrase of practical significance, as the keyword describing picture semantic feature for extracting word frequency in described residue text.
Preferably, described device also comprises:
First practical significance abstraction module, for calling the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retains the word in described residue text or phrase; Described entity word stock contains the entity word with practical significance.
Preferably, described device also comprises:
Second practical significance abstraction module, for extracting word or the phrase in described residue text with practical significance according to part of speech, described extraction process comprises: when the word in described residue text or phrase are any one in interjection, pronoun or auxiliary words of mood, abandons the word in described residue text or phrase.
Preferably, described keyword abstraction module also comprises:
Edge word statistics submodule, for the adjacent co-occurrence frequency according to other vocabulary in described keyword and described residue text, adds up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
Preferably, described device also comprises:
Filtering module, at the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtains remaining picture;
Described picture recommending module, specifically for recommending user by described residue picture.
Preferably, described picture recommending module comprises:
Number statistical submodule, for according to described search daily record, adds up described corresponding with the respective objects picture of Keywords matching on-line query request number;
Descending recommends submodule, recommends user for the descending according to on-line query request number by corresponding with the respective objects picture of Keywords matching.
Preferably, this Target Photo is the picture that the query strategy corresponding with this inquiry request mates most; Described result picture is other pictures being greater than matching threshold except Target Photo.
Compared with prior art, the application has the following advantages:
Adopt visual signature to describe inquiry picture relative to prior art, the application adopts keyword to describe the picture semantic feature of inquiry picture, and in search daily record, record Target Photo corresponding to the whole network on-line query request and corresponding keyword, picture semantic feature described by keyword can reflect the hobby of user, like this, when a submit queries request, the application can according to the keyword of Target Photo in the keyword of obtained Target Photo and described search daily record, coupling obtains having Target Photo corresponding to other user's inquiry request of same interest hobby, also namely the hobby of user can be agreed with the respective objects picture of Keywords matching, therefore, active user is recommended with the respective objects picture of Keywords matching by what extract from search daily record, provide the picture agreeing with users ' individualized requirement, extend the acquisition channel of user interest information.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of picture recommendation method in a kind of photographic search engine of prior art;
Fig. 2 is the process flow diagram of a kind of picture recommendation method embodiment of the application;
Fig. 3 is the structural drawing of a kind of picture recommendation apparatus of the application embodiment.
Embodiment
For enabling above-mentioned purpose, the feature and advantage of the application more become apparent, below in conjunction with the drawings and specific embodiments, the application is described in further detail.
Users ' individualized requirement is derived from the hobby of user often, and such as, certain user has the hobby of star-pursuing, and it is the bean vermicelli of Liu Dehua, then he is when uploading the picture of Liu Dehua, probably also wishes the picture such as film poster, individual's description seeing Liu Dehua; And for example, another user is moviegoer, and it has sincere hobby to " when happiness is knocked at the door " this film, then he is when uploading the film poster of " when happiness is knocked at the door ", probably also wishes other the different placard seeing this films more.The similar Search Results of prior art vision to meet users ' individualized requirement in the scenario above.
One of core idea of the embodiment of the present application is, the local feature inputting picture according to active user obtains Target Photo and the multiple result pictures similar or identical with its feature, the result picture place page is analyzed respectively, the Word message such as title, text in comprehensive each page, the keyword obtained associates with Target Photo, picture semantic feature described by keyword can reflect the hobby of user, like this, when a submit queries request, the application can according to the keyword of Target Photo in the keyword of obtained Target Photo and described search daily record, coupling obtains having corresponding Target Photo corresponding to other user's inquiry request of same interest hobby, also namely the hobby of user can be agreed with the respective objects picture of Keywords matching, therefore, respective objects picture with Keywords matching is recommended user the picture agreeing with users ' individualized requirement can be provided, the acquisition channel of extending user information of interest.
With reference to Fig. 2, show the process flow diagram of a kind of picture recommendation method embodiment of the application, specifically can comprise:
The inquiry request of step 201, reception user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
The application can be applied in photographic search engine, in order to expand the function of photographic search engine, also, photographic search engine is possessed original in scheme to search figure function, possesses the picture recommendation function of the application simultaneously.In fact, the application can also be applied to other search engine or searcher, and the application is not limited concrete applied environment.
In practice, user can submit on-line query request in a browser, the mode of submission on-line query request here can comprise directly uploads local picture, or the network address of picture is provided, by server automatic downloading picture, the mode of the application to concrete submission on-line query request is not limited.Also, namely, in the embodiment of the present application, the picture directly corresponding with this inquiry request can comprise the local picture that user directly uploads, and the network address that also can comprise the picture provided according to user obtains picture.
In specific implementation, server can according to the vision content of the direct corresponding picture of this inquiry request, extract local feature, then picture searching is carried out, mate with the local feature of picture each in database, if matching rate, in certain threshold range (as > 90%), can think that the vision content of the two is consistent.
For the picture directly corresponding for this inquiry request and matching result, the two only has fine distinction, the difference etc. as whether with watermark, little picture and large picture; Exclude these fine distinctions, the two is exactly identical picture.
Consider that the direct corresponding picture of this inquiry request may be the picture of the poor qualities such as the picture of band watermark or little picture, if it can be used as the storage object of search daily record, and be derived from search daily record eventually to the picture that user recommends, like this, recommend to be with the figure sector-meeting of the poor quality such as watermark or little picture to affect the search experience of user to user.Therefore, in a preferred embodiment of the present application, the picture that the query strategy corresponding with inquiry request is mated most as Target Photo, and using this Target Photo as search daily record storage object.In practice, mate database used and often store some and be not with watermark and larger-size picture, like this, to recommend not with watermark and larger-size picture can improve the search experience of user to user.
In a preferred embodiment of the present application, result picture is other pictures being greater than matching threshold in database except Target Photo, and the degree that conforms to of the query strategy that namely result picture is corresponding with inquiry request is less than the degree that conforms to of the Target Photo query strategy corresponding with inquiry request.In the present embodiment, the Target Photo obtained and result picture sort by matching degree, and the picture mated most with inquiry request is Target Photo, and remaining picture as a result picture carries out sequence displaying by matching degree.In other embodiments, the corresponding result of the inquiry request of user can sort by picture size or issuing time, the picture of size is maximum or nearest issue as Target Photo, remaining picture as a result picture by size from large to small or issuing time by near to far carrying out sequence displaying.Under normal conditions, result picture and Target Photo only have fine distinction, the difference etc. as whether with watermark, little picture and large picture; Exclude these fine distinctions, the two is exactly identical picture.
Be appreciated that when the application is applied to photographic search engine, described result picture can also be returned to user as Search Results by server, to meet some search need of user.Such as certain user likes collection picture, and least patient is exactly have watermark above beautiful figure, as long as uploading pictures is to photographic search engine, and the picture just can found not with watermark of clicking; And for example, little picture can be uploaded, search out each version of this little picture, as clear large figure etc.
In a kind of application example of the application, the vision content of the described picture directly corresponding according to this inquiry request, the step extracting local feature specifically can comprise:
First, the size of the direct corresponding picture of this inquiry request is normalized, oversize or too small picture is transformed within 640*640 ~ 300*300; Then the picture after two-dimentional local feature monitoring matrix and normalization is used to carry out convolution operation; Moreover Scan orientation goes out the position that local extremum (maxima and minima) is wherein put in the picture after convolution; Finally, according to the comparison of light and shade of Local Extremum near zone, extract the local feature of the directly corresponding picture of this inquiry request.It should be noted that, in order to realize mating object, the direct corresponding picture of this inquiry request should be consistent with the picture size after normalization in database with identical original size with it, such as, is all 300*300.
With reference to table 1, show the dimension of picture signal before and after a kind of normalization of the application.
Table 1
In other embodiments, described result picture also for carry out feature extraction to Target Photo, can be searched in a database, carries out mating obtained picture with the local feature of picture each in database.
Step 202, web page text according to described result picture place, extract the keyword describing picture semantic feature, as the keyword of this Target Photo;
Because result picture comes from network, therefore can record the web page text of each width result picture in the database of search engine or searcher, these web page texts generally include the text message of webpage, as page title, and the description text etc. of picture periphery.
Due to the picture that result picture is identical or approximate with Target Photo, under normal conditions, the two only has fine distinction, the difference etc. as whether with watermark, little picture and large picture; Exclude these fine distinctions, the two is exactly identical picture, and that is, result picture can represent Target Photo completely.
Like this, according to the web page text of result picture, the keyword extracted can the picture semantic feature of objective description Target Photo, and the picture semantic feature of Target Photo can reflect the hobby of user to a certain extent, such as, user search obtains the picture of Liu Dehua, this user of very possible explanation is the bean vermicelli of Liu Dehua, and for example, the film poster that user search obtains " when happiness is knocked at the door ", very possible this user of explanation is the fan of " when happiness is knocked at the door " etc.
In a preferred embodiment of the present application, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, may further include:
Sub-step A1, cluster analysis is carried out to described web page text;
Sub-step A2, foundation cluster analysis result, remove web page text isolated in described web page text, obtains remaining text;
In specific implementation, the web page text of every width result picture can be considered as a document, carry out cluster analysis to the web page text of all result pictures, the isolated text those do not flocked together is considered as noise removal and falls.The principle of cluster analysis is arest neighbors binary tree cluster, when being applied to web page text, it is according to the repetition degree of web page text, be considered as a class merge repeating two parts of maximum web page texts, and the class after merging is considered as a web page text, iteration repeats down, till the repetition degree between two maximum web page texts that repeats can not reach merging threshold value.
With reference to table 2 and table 3, remain the example of text after respectively illustrating a kind of original web page text of the application and cluster analysis, wherein, original web page text comprises text corresponding to 1-9 nine parts of webpages, cluster analysis eliminate wherein be numbered 2,4,9 noise texts, obtain remain text.
Table 2
Table 3
In the ideal case, the web page text of result picture can describe the semantic content of corresponding picture truely and accurately, but because the quality of web page text is uneven, at some in particular cases, the semantic content of web page text and picture is also uncorrelated.Such as, in table 2 original web page text 2,9.Although (text 4 is relevant to picture semantic content, does not reach merging threshold value, be therefore also removed with the repetition degree of other texts.)
In practice, above-mentioned result picture is ideally in the great majority, and result picture is in particular cases very indivedual, like this, when cluster analysis, the web page text of result picture ideally can flock together, and the web page text of result picture is in particular cases isolated; Therefore, what those can not flock together by above-mentioned cluster analysis be considered as noise removal with the object in Target Photo or the incoherent isolated text of scene falls, to improve the accuracy of keyword abstraction.
Sub-step A3, to extract word frequency in described residue text the highest and have word or the phrase of practical significance, as the keyword describing picture semantic feature.
The application can provide following extraction in described residue text to have the word of practical significance or the scheme of phrase:
Scheme one,
Word or the phrase in described residue text with practical significance can be extracted by following steps:
According to the entity dictionary constructed in advance, extract word or the phrase in described residue text with practical significance, described entity word stock contains the entity word with practical significance, and described extraction process can comprise:
Call the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retain the word in described residue text or phrase.
Here entity word mainly refers to the word representing single or multiple entitative concept, and it mainly comprises and is once called as noun, as name, movie name, item name etc.In practice, can collect the entity word under hobby classification, and construct corresponding entity dictionary in advance, hobby classification here both can comprise; The amusement classification such as film, TV, star, music, animation, also can the leisure such as books, electronic product, clothes, shoes and hats classification etc.The make of the application to concrete hobby classification and entity dictionary is not limited.
Scheme two,
Word or the phrase in described residue text with practical significance can be extracted by following steps:
Extract word or the phrase in described residue text with practical significance according to part of speech, described extraction process specifically can comprise:
When word in described residue text or phrase are any one in interjection, pronoun or auxiliary words of mood, abandon the word in described residue text or phrase.
Because interjection, pronoun or auxiliary words of mood etc. are everyday words, usually there is no practical significance, therefore when extracting, discard processing can be carried out to it.It should be noted that, except interjection, pronoun or auxiliary words of mood, this programme can also according to actual conditions, abandon word or the phrase of other part of speech in described residue text, as any one in adverbial word, preposition, conjunction, structural auxiliary word, dynamically auxiliary word, onomatopoeia etc., the application is not limited the part of speech specifically abandoned.
It should be noted that, in order to alleviate to extract in described residue text, there is the word of practical significance or the workload of phrase, in the embodiment of the present application, preferably, first word that in described residue text, word frequency is the highest can be extracted or phrase is tentatively extracted result, then, from described preliminary extraction result, extract word or the phrase with practical significance, finally extracted result.Certainly, first those skilled in the art also as required, can extract word or the phrase in described residue text with practical significance, and then extract the highest word of word frequency or phrase, the application is not limited concrete succession.
In addition, above-mentioned two kinds of schemes extracting the word or phrase having practical significance in described residue text can be used alone or are combined, or, those skilled in the art can also be according to actual needs, adopt other to extract in described residue text and have the word of practical significance or the scheme of phrase, the application is not limited this.
In another preferred embodiment of the present application, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, can also comprise:
According to the adjacent co-occurrence frequency of other vocabulary in described keyword and described residue text, add up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
Suppose that user uploads the picture of Liu Dehua, and step 201-203 extracts keyword---" Liu Dehua " that obtain describing picture semantic feature; In fact this user also wishes the picture such as film poster, individual's description seeing Liu Dehua, so, can " Liu Dehua " be keyword, add up other vocabulary that in described residue text, co-occurrence number of times adjacent with " Liu Dehua " is more, as " film ", " classic film ", " description " etc., like this, the keyword finally obtained can comprise: " Liu De China film ", " Liu De China classic film ", " Liu Dehua description " etc.
Step 203, in search daily record, carry out the coupling of keyword, and the respective objects picture with Keywords matching is recommended user; The Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword.
Network operating system is designed with various journal file usually; as application log; security log, system journal etc.; when user carries out certain operations in network system; these journal files can record some related contents of operation usually; IP (agreement interconnected between network, Internet Protocol), time, user name etc. as used in user.
The search daily record of the application generates for the on-line query request of the whole network user, with prior art unlike, Target Photo corresponding to this on-line query request and corresponding keyword can be recorded in described search daily record, wherein, described keyword obtains by performing step 201-202.Here the whole network user can comprise the user of internet, namely also when the user of internet submits on-line query request in search engine or searcher, the server of search engine or searcher can generate searches for daily record accordingly, and the application can search for daily record by the collects of all search engines or searcher from internet, obtain searching for daily record.The application only specifies the storage content of search daily record, and can not be limited the obtain manner of concrete collection mode or search daily record.
In specific implementation, the search daily record of coupling institute foundation should be the search daily record of the whole network user, inquire about obtain corresponding Target Photo to inquire other users matched with the keyword of this Target Photo, here Keywords matching mainly refers to the keyword that the keyword searching for the Target Photo recorded in daily record is identical with the keyword of current goal picture, comprise this Target Photo, or overlap each other, etc.
The picture recommendation function that the application provides can meet users ' individualized requirement preferably, because the picture semantic feature in the application described by keyword can reflect the hobby of user, like this, when a submit queries request, the application can according to the keyword of Target Photo in the keyword of obtained Target Photo and described search daily record, other user that coupling obtains having same interest hobby inquires about the corresponding Target Photo obtained, and also namely can agree with the hobby of user with the respective objects picture of Keywords matching.
In a preferred embodiment of the present application, before the respective objects picture with Keywords matching is recommended user, described method can also comprise:
At the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtain remaining picture;
The described step respective objects picture with Keywords matching being recommended user can be that described residue picture is recommended user.
Mention above, in the ideal case, in each Target Photo corresponding with keyword, it is identical or approximate that two identical or approximate width pictures only have the nuances such as difference whether with watermark, little picture and large picture usually; In addition, be obtain according to the Keywords matching describing picture semantic feature to corresponding each Target Photo of Keywords matching; Therefore, can think, if the pictures identical or approximate with there are more than two width or two pairs in corresponding each Target Photo of Keywords matching, then the meaning of not recommending, therefore filtering is carried out to it.
In another preferred embodiment of the present application, the described step respective objects picture with Keywords matching being recommended user, may further include:
According to described search daily record, the on-line query request number that statistics is described corresponding with the respective objects picture of Keywords matching;
User is recommended corresponding with the respective objects picture of Keywords matching according to the descending of on-line query request number.
In some cases, described may be big figure to the number of corresponding each Target Photo of Keywords matching, more than 100 width, whether the picture of these big figures agrees with users ' individualized requirement is difficult to expect, and need point multipage to be shown in a browser by the picture of these big figures, make user need the content extracted from multipage required for oneself.
This preferred embodiment recommends user by corresponding with the respective objects picture of Keywords matching according to the descending of on-line query request number, on-line query request number more the bright corresponding picture of multilist more by had same interest hobby user pay close attention to, also be, the application can the high picture of the many attention rates of preferential recommendation on-line query request number, therefore, the picture of preferential recommendation can agree with users ' individualized requirement better, the experience of adding users.
The application can provide the application example in following scene:
Application example 1,
Step B1, receive the picture of Liu Dehua that user uploads, and carry out searching for this picture and obtain Target Photo corresponding to this picture and the result picture identical or approximate with Target Photo;
Step B2, web page text according to described result picture place, extract the keyword describing picture semantic feature, as the keyword of the picture of Liu Dehua, and such as " Liu De China film ", " Liu De China classic film ", " Liu Dehua description " etc.;
Step B3, in search daily record, carry out the coupling of keyword, obtain Target Photo corresponding to other reflection hobbies liking the user of Liu De China to upload equally (picture concerned as the more Liu De China such as film poster, individual's description of Liu Dehua), and recommend user.
Application example 2,
" failing in love 33 days " film poster that step C1, reception user upload; Corresponding Target Photo and the result picture identical or approximate with Target Photo is obtained by the search of this film poster;
Step C2, web page text according to described result picture place, extract the keyword describing picture semantic feature, as the keyword of this Target Photo, as " failing in love 33 days " etc.;
Step C3, in search daily record, carry out the coupling of keyword, obtain the Target Photo (the different placards of such as this film) that other reflection hobbies liking the user of " failing in love 33 days " this film to upload are corresponding, and recommend user.
For making those skilled in the art understand the application better, the process flow diagram of the method example of recommending star's picture in a kind of photographic search engine of the application being below provided, specifically can comprising:
Step 1, receiving the girl of golden hair that user uploads can the graceful photo in Jede;
Step 2, can extract visual signature the graceful photo in Jede from the girl of golden hair, compare with the visual signature of picture in database, obtaining can the consistent Target Photo of the vision content of the graceful photo in Jede and result picture with the girl of golden hair;
Step 3, carry out cluster analysis to the web page text of result picture, the isolated text those do not flocked together is considered as noise removal and falls, and extracts word frequency in residue text the highest and have word or the phrase of practical significance, as keyword;
Such as, table 4 shows a kind of word frequency example remained in text of the application.
Table 4
| Word | Word frequency in text |
| They | 12 |
| Girl can Jede graceful | 10 |
| Film festival | 5 |
| Venice | 4 |
| ... | ... |
Wherein, " they " that word frequency is the highest do not have practical significance, and the keyword therefore finally obtained is " girl can Jede graceful ".
Target Photo corresponding to on-line query request and corresponding keyword that the whole network user submits to is recorded in the search daily record of step 4, photographic search engine;
Step 5, be drawn into Target Photo keyword after, whether by Keywords matching (identical, comprise mutually or overlapping to some extent), inquiry obtains the Target Photo having semantic association with each keyword;
Step 6, after filtering out the identical or approximate Target Photo of Target Photo corresponding with current queries in Query Result, remain the on-line query request number of picture in statistics search daily record, and part pictures maximum for on-line query request number is recommended user.
Such as, when keyword is " girl can Jede graceful ", the keyword be associated with " girl can Jede graceful " in search daily record and the on-line query request number of the corresponding picture of these keywords as shown in table 4,
Table 4
| Keyword | Corresponding Target Photo | Corresponding on-line query request number |
| Beauty girl can Jede graceful | Picture a | 22 |
| Girl can Jede graceful | Picture b | 19 |
| Girl can the graceful stage photo in Jede | Picture c | 8 |
| ... | ... | ... |
If for user recommends 2 width pictures, so the application will recommend picture a and picture b.
First, said method example can extract keyword to describe the picture semantic feature of Target Photo, as the name of star in photo by photographic search engine.
Secondly, said method example can provide the picture with identical semantic feature as content recommendation for user, also namely can provide the picture materials agreeing with users ' individualized requirement.Like this, when Target Photo be golden hair girl can the graceful photo in Jede time, the different pictures about this star can be recommended, as the photo of these other hair colors of star; Instead of the picture only recommending vision similar.
Corresponding to preceding method embodiment, present invention also provides a kind of picture recommendation apparatus, with reference to Fig. 3, specifically can comprise:
Picture searching module 301, for receiving the inquiry request of user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
Keyword abstraction module 302, for the web page text according to described result picture place, extracts the keyword describing picture semantic feature, as the keyword of this Target Photo;
Matching module 303, for carrying out the coupling of keyword in search daily record; The Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword; And
Picture recommending module 304, for recommending user by the respective objects picture with Keywords matching.
In the embodiment of the present application, preferably, this Target Photo is the picture that the query strategy corresponding with this inquiry request mates most; Described result picture is other pictures being greater than matching threshold in the database of server end except Target Photo.
In a preferred embodiment of the present application, described keyword abstraction module 302 may further include:
Remove submodule, for according to the result of described web page text being carried out to cluster analysis, remove web page text isolated in described web page text, obtain remaining text; And
Extract submodule, the highest and there is word or the phrase of practical significance, as the keyword describing picture semantic feature for extracting word frequency in described residue text.
In another preferred embodiment of the present application, described device can also comprise:
First practical significance abstraction module, for the entity dictionary that foundation constructs in advance, extract word or the phrase in described residue text with practical significance, described entity word stock contains the entity word with practical significance, described extraction process comprises: when the entity word in the word in described residue text or phrase and described entity dictionary matches, retain the word in described residue text or phrase.
In another preferred embodiment of the application, described device can also comprise:
Second practical significance abstraction module, for calling the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retains the word in described residue text or phrase; Described entity word stock contains the entity word with practical significance.
In a preferred embodiment of the present application, described keyword abstraction module 302 can also comprise:
Edge word statistics submodule, for the adjacent co-occurrence frequency according to other vocabulary in described keyword and described residue text, adds up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
In the embodiment of the present application, preferably, described device can also comprise:
Filtering module, at the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtains remaining picture;
Now, described picture recommending module 304, can specifically for recommending user by described residue picture.
In the embodiment of the present application, preferably, described picture recommending module 304 specifically can comprise:
Number statistical submodule, for according to described search daily record, adds up described corresponding with the respective objects picture of Keywords matching on-line query request number; And
Descending recommends submodule, recommends user for the descending according to on-line query request number by corresponding with the respective objects picture of Keywords matching.
For device embodiment, due to itself and embodiment of the method basic simlarity, so description is fairly simple, relevant part illustrates see the part of embodiment of the method.
Each embodiment in this instructions all adopts the mode of going forward one by one to describe, and what each embodiment stressed is the difference with other embodiments, between each embodiment identical similar part mutually see.
A kind of picture recommendation method above the application provided and device, be described in detail, apply specific case herein to set forth the principle of the application and embodiment, the explanation of above embodiment is just for helping method and the core concept thereof of understanding the application; Meanwhile, for one of ordinary skill in the art, according to the thought of the application, all will change in specific embodiments and applications, in sum, this description should not be construed as the restriction to the application.
Claims (16)
1. a picture recommendation method, is characterized in that, comprising:
Receive the inquiry request of user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
According to the web page text at described result picture place, extract the keyword describing picture semantic feature, as the keyword of this Target Photo;
In search daily record, carry out the coupling of keyword, and the respective objects picture with Keywords matching is recommended user; Wherein, described respective objects picture is have other user that same interest likes to inquire about with described user the corresponding Target Photo obtained, the Target Photo that described search log recording has the on-line query request of the whole network user corresponding and corresponding keyword.
2. the method for claim 1, is characterized in that, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, comprising:
According to the result of described web page text being carried out to cluster analysis, remove web page text isolated in described web page text, obtain remaining text;
Extract word frequency in described residue text the highest and there is word or the phrase of practical significance, as the keyword describing picture semantic feature.
3. method as claimed in claim 2, be is characterized in that, extracted word or the phrase in described residue text with practical significance by following steps:
Call the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retain institute's predicate or phrase; Described entity word stock contains the entity word with practical significance.
4. method as claimed in claim 2, be is characterized in that, extracted word or the phrase in described residue text with practical significance by following steps:
Extract word or the phrase in described residue text with practical significance according to part of speech, described extraction process comprises:
When word in described residue text or phrase are any one in interjection, pronoun or auxiliary words of mood, abandon institute's predicate or phrase.
5. method as claimed in claim 2, is characterized in that, the described web page text according to described result picture place, extracts the step of the keyword describing picture semantic feature, also comprises:
According to the adjacent co-occurrence frequency of other vocabulary in described keyword and described residue text, add up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
6. the method according to any one of claim 1 to 5, is characterized in that, described method also comprises:
At the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtain remaining picture;
The described step respective objects picture with Keywords matching being recommended user is that described residue picture is recommended user.
7. the method according to any one of claim 1 to 5, is characterized in that, the described step respective objects picture with Keywords matching being recommended user, comprising:
According to described search daily record, the on-line query request number that statistics is described corresponding with the respective objects picture of Keywords matching;
According to the descending of on-line query request number, the respective objects picture with Keywords matching is recommended user.
8. the method according to any one of claim 1 to 5, is characterized in that, this Target Photo is the picture that the query strategy corresponding with this inquiry request mates most; Described result picture is other pictures being greater than matching threshold except Target Photo.
9. a picture recommendation apparatus, is characterized in that, comprising:
Picture searching module, for receiving the inquiry request of user, and search obtains the Target Photo corresponding with this inquiry request and the result picture identical or approximate with this Target Photo;
Keyword abstraction module, for the web page text according to described result picture place, extracts the keyword describing picture semantic feature, as the keyword of this Target Photo;
Matching module, for carrying out the coupling of keyword in search daily record; Wherein, the described search log recording Target Photo that has the on-line query request of the whole network user corresponding and corresponding keyword; And
Picture recommending module, for recommending user by the respective objects picture with Keywords matching; Wherein, described respective objects picture is have other user that same interest likes to inquire about with described user the corresponding Target Photo obtained.
10. device as claimed in claim 9, it is characterized in that, described keyword abstraction module comprises:
Remove submodule, for according to the result of described web page text being carried out to cluster analysis, remove web page text isolated in described web page text, obtain remaining text; And
Extract submodule, the highest and there is word or the phrase of practical significance, as the keyword describing picture semantic feature for extracting word frequency in described residue text.
11. devices as claimed in claim 10, is characterized in that, also comprise:
First practical significance abstraction module, for calling the entity dictionary constructed in advance, when the entity word in the word in described residue text or phrase and described entity dictionary matches, retains the word in described residue text or phrase; Described entity word stock contains the entity word with practical significance.
12. devices as claimed in claim 10, is characterized in that, also comprise:
Second practical significance abstraction module, for extracting word or the phrase in described residue text with practical significance according to part of speech, described extraction process comprises: when the word in described residue text or phrase are any one in interjection, pronoun or auxiliary words of mood, abandons the word in described residue text or phrase.
13. devices as claimed in claim 10, it is characterized in that, described keyword abstraction module also comprises:
Edge word statistics submodule, for the adjacent co-occurrence frequency according to other vocabulary in described keyword and described residue text, adds up edge word adjacent with described keyword in described residue text; Using described edge word keyword as description picture semantic feature together with keyword.
14. devices according to any one of claim 9 to 13, is characterized in that, also comprise:
Filtering module, at the picture identical or approximate with filtering in corresponding each Target Photo of Keywords matching, obtains remaining picture;
Described picture recommending module, specifically for recommending user by described residue picture.
15. devices according to any one of claim 9 to 13, it is characterized in that, described picture recommending module comprises:
Number statistical submodule, for according to described search daily record, adds up described corresponding with the respective objects picture of Keywords matching on-line query request number;
Descending recommends submodule, recommends user for the descending according to on-line query request number by corresponding with the respective objects picture of Keywords matching.
16. devices according to any one of claim 9 to 13, it is characterized in that, this Target Photo is the picture that the query strategy corresponding with this inquiry request mates most; Described result picture is other pictures being greater than matching threshold except Target Photo.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210009043.7A CN102609458B (en) | 2012-01-12 | 2012-01-12 | A kind of picture recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210009043.7A CN102609458B (en) | 2012-01-12 | 2012-01-12 | A kind of picture recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102609458A CN102609458A (en) | 2012-07-25 |
| CN102609458B true CN102609458B (en) | 2015-08-05 |
Family
ID=46526830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210009043.7A Active CN102609458B (en) | 2012-01-12 | 2012-01-12 | A kind of picture recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102609458B (en) |
Families Citing this family (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831176B (en) * | 2012-07-30 | 2016-12-21 | 东莞宇龙通信科技有限公司 | The method of commending friends and server |
| CN102902771A (en) * | 2012-09-27 | 2013-01-30 | 百度国际科技（深圳）有限公司 | Method, device and server for searching pictures |
| CN103064903B (en) * | 2012-12-18 | 2017-08-01 | 厦门市美亚柏科信息股份有限公司 | Picture retrieval method and device |
| CN103886506B (en) * | 2012-12-20 | 2018-08-10 | 联想(北京)有限公司 | A kind of information processing method and electronic equipment |
| CN103902678A (en) * | 2014-03-21 | 2014-07-02 | 百度在线网络技术（北京）有限公司 | Search recommendation method and device |
| CN104090929A (en) * | 2014-06-23 | 2014-10-08 | 吕志雪 | Recommendation method and device of personalized picture |
| CN104156694B (en) * | 2014-07-18 | 2019-03-19 | 百度在线网络技术（北京）有限公司 | A kind of method and apparatus of target object in image for identification |
| CN104239457B (en) * | 2014-09-02 | 2019-03-12 | 百度在线网络技术（北京）有限公司 | Search for recommended method and device |
| CN104504111B (en) * | 2014-12-30 | 2018-12-21 | 百度在线网络技术（北京）有限公司 | The recommended method and device of picture material |
| CN104504110B (en) | 2014-12-30 | 2019-05-03 | 百度在线网络技术（北京）有限公司 | Searching method and device |
| CN104881451A (en) * | 2015-05-18 | 2015-09-02 | 百度在线网络技术（北京）有限公司 | Image searching method and image searching device |
| CN104994404A (en) * | 2015-07-06 | 2015-10-21 | 无锡天脉聚源传媒科技有限公司 | Method and device for obtaining keywords for video |
| CN105159938B (en) * | 2015-08-03 | 2018-11-30 | 百度在线网络技术（北京）有限公司 | Search method and device |
| US10063751B2 (en) * | 2015-09-24 | 2018-08-28 | Qualcomm Incorporated | System and method for accessing images with a captured query image |
| CN107305566B (en) * | 2016-04-21 | 2019-10-18 | 北京搜狗科技发展有限公司 | A kind of method and device to search for information matches picture |
| CN108230220B (en) * | 2016-12-15 | 2022-04-26 | 百度在线网络技术（北京）有限公司 | Watermark adding method and device |
| CN106791912B (en) * | 2016-12-28 | 2020-07-03 | Tcl科技集团股份有限公司 | Poster picture matching method and system |
| CN107423318A (en) * | 2017-03-27 | 2017-12-01 | 北京珠穆朗玛移动通信有限公司 | A kind of method and mobile terminal of picture mark |
| CN107862561A (en) * | 2017-09-15 | 2018-03-30 | 广州唯品会研究院有限公司 | A kind of method and apparatus that user-interest library is established based on picture attribute extraction |
| CN107909088B (en) * | 2017-09-27 | 2022-06-28 | 百度在线网络技术（北京）有限公司 | Method, apparatus, device and computer storage medium for obtaining training samples |
| CN110019675B (en) * | 2017-12-01 | 2021-10-15 | 北京搜狗科技发展有限公司 | Keyword extraction method and device |
| WO2019150416A1 (en) * | 2018-01-30 | 2019-08-08 | 株式会社オプティム | Electronic data conversion system, method and program |
| CN110134850B (en) * | 2018-02-09 | 2024-05-14 | 北京搜狗科技发展有限公司 | Searching method and device |
| CN110309324B (en) * | 2018-03-09 | 2024-03-22 | 北京搜狗科技发展有限公司 | Searching method and related device |
| CN111859095A (en) * | 2019-04-02 | 2020-10-30 | 搜狗（杭州）智能科技有限公司 | Picture identification method and device |
| CN110941766B (en) * | 2019-12-10 | 2023-10-20 | 北京字节跳动网络技术有限公司 | Information pushing method, device, computer equipment and storage medium |
| CN113590861B (en) * | 2020-04-30 | 2024-06-18 | 北京搜狗科技发展有限公司 | Picture information processing method and device and electronic equipment |
| CN113590862A (en) * | 2020-04-30 | 2021-11-02 | 北京搜狗科技发展有限公司 | Picture information processing method and device and electronic equipment |
| CN111831847B (en) * | 2020-07-08 | 2023-08-18 | 嘉兴智设信息科技有限公司 | Similar picture set recommendation method and system |
| CN113468414A (en) * | 2021-06-07 | 2021-10-01 | 广州华多网络科技有限公司 | Commodity searching method and device, computer equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241512A (en) * | 2008-03-10 | 2008-08-13 | 北京搜狗科技发展有限公司 | Search method for redefining enquiry word and device therefor |
| CN102270234A (en) * | 2011-08-01 | 2011-12-07 | 北京航空航天大学 | Image search method and search engine |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7099860B1 (en) * | 2000-10-30 | 2006-08-29 | Microsoft Corporation | Image retrieval systems and methods with semantic and feature based relevance feedback |
-
2012
- 2012-01-12 CN CN201210009043.7A patent/CN102609458B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241512A (en) * | 2008-03-10 | 2008-08-13 | 北京搜狗科技发展有限公司 | Search method for redefining enquiry word and device therefor |
| CN102270234A (en) * | 2011-08-01 | 2011-12-07 | 北京航空航天大学 | Image search method and search engine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102609458A (en) | 2012-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102609458B (en) | A kind of picture recommendation method and device | |
| US11238066B2 (en) | Generating personalized clusters of multimedia content elements based on user interests | |
| US20160034514A1 (en) | Providing search results based on an identified user interest and relevance matching | |
| Yang et al. | Crowdsourced time-sync video tagging using semantic association graph | |
| CN106294358A (en) | The search method of a kind of information and system | |
| JP7395377B2 (en) | Content search methods, devices, equipment, and storage media | |
| Ding et al. | Hierarchical clustering-based navigation of image search results | |
| Figueiredo et al. | Evidence of quality of textual features on the web 2.0 | |
| Shah et al. | A review: Web content mining techniques | |
| Elbassuoni et al. | ROXXI: Reviving witness dOcuments to eXplore eXtracted Information | |
| CN108509449B (en) | Information processing method and server | |
| Waitelonis et al. | Use what you have: Yovisto video search engine takes a semantic turn | |
| Zhao et al. | A search result ranking algorithm based on web pages and tags clustering | |
| Vallet et al. | Exploiting external knowledge to improve video retrieval | |
| Qian et al. | Image retrieval by user-oriented ranking | |
| Choi et al. | Consento: a new framework for opinion based entity search and summarization | |
| Bjelland et al. | Web link analysis: estimating document’s importance from its context | |
| Tsapatsoulis | Web image indexing using WICE and a learning-free language model | |
| Vallet et al. | Exploiting semantics on external resources to gather visual examples for video retrieval | |
| Dinesh | Real world evaluation of approaches to research paper recommendation | |
| Dai et al. | Efficient tag mining via mixture modeling for real-time search-based image annotation | |
| LAIC et al. | Automatic image annotation | |
| Zhou et al. | Automatic image annotation by using relevant keywords extracted from auxiliary text documents | |
| Bhargav | Efficient features for movie recommendation systems | |
| Zinger et al. | Clustering and semantically filtering web images to create a large-scale image ontology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |