JP5296014B2 - Search device, method and program - Google Patents

Description

  The present invention relates to web search.

  2. Description of the Related Art Conventionally, in the field of so-called web search, various approaches have been made to present a search result that matches a searcher's search intention. For example, in Patent Document 1, in order to support a search without omission, one or more search keywords (referred to as a “search query”) input by a user in a search are extended to other related synonyms. To search.

Japanese Patent Application Laid-Open No. 5-181912 JP 2002-288201 A

  However, with respect to the above-mentioned Patent Document 1, although the search range is expanded, the search is not expanded considering the search intention such as what the user wants to know about the search query, but the search suitable for the search intention is realized. It was not a thing.

  In terms of consideration of search intention, the user can input a long search sentence, understand the user's search intention by morphological analysis of the search sentence, and support the creation of an appropriate search expression. There is also a proposal for adjusting the amount, content, etc. (see, for example, Patent Document 2), but in this case, there is a problem that an excessive user load of inputting a long sentence occurs.

  In view of the above problems, an object of the present invention is to provide a search result suitable for a user's search intention without excessive user load.

Based on the above object, a search device according to one aspect (1) of the present invention represents search query acquisition means for acquiring a search query from a terminal, and attributes related to the search query based on the acquired search query. Attribute candidate generation means for generating an attribute candidate word that is a word candidate, and word processing based on a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms for the generated attribute candidate word Pattern processing means for extracting the attribute by applying, action candidate generation means for generating an action candidate corresponding to the extracted attribute, and a search character string corresponding to the generated action candidate Search is performed, and outputs the additional keywords or images based on the search result to the terminal with a candidate of the generated action action A candidate output means, if the candidate of one of the actions of the candidate of the action is specified, and perform a search using a search character string corresponding to the action which is the designated, the additional keyword or image In the case of being specified, it has a search means for executing a search using the specified additional keyword or image, and a result output means for outputting the result of the executed search to the terminal .

The search method according to another aspect ( 5 ) of the present invention is the above-described aspect captured in the category of the method. The computer acquires a search query from a terminal , and the computer acquires the search query. Based on the search query, an attribute candidate generation step of generating an attribute candidate word that is a word candidate representing an attribute related to the search query, and the computer removes a predetermined type of word from the generated attribute candidate word and A pattern processing step for extracting the attribute by performing word processing with a predetermined pattern including at least one of synonym unification, and an action candidate generation step for generating a candidate for an action corresponding to the extracted attribute If, test the computer, using the search character string corresponding to the generated candidate of the action Was carried out, and actions candidate output step of outputting an additional keyword or image based on the search result to the terminal with a candidate of the generated action, computer, designated candidate of any actions of the candidate of the action When the search is performed, a search is performed using the search character string corresponding to the specified action, and when the additional keyword or image is specified, the search is performed using the specified additional keyword or image. And a result output step in which a computer outputs a result of the executed search to the terminal .

A search program according to another aspect ( 6 ) of the present invention is the one obtained by capturing the above aspect in a category of a computer program. By controlling the computer, a search query is acquired from a terminal , and the acquired search query Based on the above, a candidate attribute word that is a candidate word representing an attribute related to the search query is generated, and the generated attribute candidate word includes a predetermined type of word removal and synonym unification. The attribute is extracted by performing word processing based on the pattern of the above, an action candidate corresponding to the extracted attribute is generated, and a search is performed using the search character string corresponding to the generated action candidate , together with the candidate actions that the additional keyword or image is the product based on the search result is output to the terminal, before When any of the action candidates is designated, a search is performed using a search character string corresponding to the designated action, and when the additional keyword or image is designated, The search is executed using the specified additional keyword or image, and the result of the executed search is output to the terminal .

In this way, based on the search query, an attribute candidate word that is a candidate word representing the attribute of the search intention is extracted, and word processing such as removing a predetermined type of word is performed on the attribute candidate word, Extracting search intention attributes in the form of remaining candidate words and their combinations, etc., and presenting the action candidates according to the attributes to the user, such as switching search types and conditions according to candidate selection Since the search intention attribute can be reflected in the search, a search result suitable for the user's search intention can be provided without excessive user load such as inputting a long sentence. The “action” is a search intention represented by a character string using a plurality of words.
In addition, search candidates that match the user's search intention are created by presenting the user with action candidates according to the search intention attribute extracted from the search query, and generating an appropriate search character string according to the selection from the candidate. Since the result can be displayed, it is possible to more effectively provide a search result suitable for the search intention of the user without excessive user load.
Also, if you search in advance for each action candidate with the search character string and output additional keywords such as more specific options together with the candidate based on the search result, it becomes easier to narrow down the information to be acquired, Search usability is improved.
In addition, when a search is performed in advance using a search character string for each candidate of an action and an image such as a product photo for each product is output together with the candidate based on the search result, the content of information provided when the candidate is specified is It is easy to grasp from the image at first glance, and the usability of search is improved.

According to another aspect (2) of the present invention, in the aspect (1) , when the additional keyword is specified, the search means is a search character string corresponding to the specified additional keyword and the action candidate. A search is performed using

  According to another aspect (3) of the present invention, in any one of the above aspects, the attribute candidate generation unit includes a related word storage unit that stores a plurality of words used simultaneously in the search query in association with each other as related words. The means extracts, as the attribute candidate word, a word associated with the input search query and the related word from the related word storage means, and the pattern processing means performs the attribute candidate word as the word processing. A proper noun removing means for removing a proper noun from the word, a noun adjective removing means for removing a nounative adjective from the attribute candidate word as the word processing, and a synonym for the attribute candidate word as the word processing Synonym processing means, and attribute candidates remaining as a result of the removal and unification are defined as the attributes.

  In this way, based on the search query, related words that are often combined are attribute candidate words, and proper nouns, nounative adjectives, and synonyms are excluded, and the remaining ones are efficiently searched using simple search attributes. The attribute of the intention can be extracted and a search result suitable for the user's search intention can be provided.

According to another aspect ( 4 ) of the present invention, in any one of the above aspects, the action candidate output unit performs a search using the search character string corresponding to the action candidate, and the search result includes a predetermined number. Fewer action candidates are excluded from display targets.

  As described above, when there is no search result for an action, the action is excluded from the display target, so that a useless operation that the search result of the selected candidate is 0 can be avoided, and the usability of the search is improved.

  It should be noted that a category (method for the apparatus, program for the method, etc.) different from each of the above-described modes and more specific modes described below are also included in the present invention. For different categories, “means” shall be appropriately read as “step”.

  According to the present invention, it is possible to provide a search result suitable for a user's search intention without excessive user load.

The functional block diagram which shows the structure of embodiment of this invention. The figure which illustrates the information (data) used by embodiment of this invention. The figure which illustrates the information (data) used by embodiment of this invention. The flowchart which shows the process sequence in embodiment of this invention. The conceptual diagram which shows the example which extracts an attribute word from an attribute candidate word in embodiment of this invention. The figure which shows an example of the screen display in embodiment of this invention. The figure which shows an example of the screen display in embodiment of this invention. The figure which shows an example of the screen display in embodiment of this invention. The figure which shows an example of the screen display in embodiment of this invention. The figure which shows an example of the screen display in embodiment of this invention.

  Next, modes for carrying out the present invention (referred to as “embodiments”) will be described with reference to the drawings. It should be noted that assumptions common to those already described in the background art and problems are omitted as appropriate.

[1. Constitution〕
In the present embodiment, as shown in FIG. 1, a search device that transmits a web search result to a terminal T in response to a search request transmitted from the terminal T via a communication network N (Internet, mobile phone network, LAN, etc.). 1 (hereinafter referred to as “this device” or “this device 1”).

  The present apparatus 1 has a general computer configuration of at least an arithmetic control unit 6 such as a CPU, a storage device 7 such as an external storage device (HDD or the like) or a main memory, and a communication means 8 (LAN) with a communication network N. Adapter, communication gateway, etc.). A terminal T is a client device used by a user, such as a personal computer (PC), a smart phone, or a mobile phone terminal device.

  In the present apparatus 1, a predetermined computer program (not shown) stored (installed) in advance in the storage device 7 controls the arithmetic control unit 6 so that the elements (10, 20, 25, etc.) shown in FIG. Etc.). Among these elements, the information storage means can be realized in the storage device 7 in any format such as various files and databases (also referred to as “DB”), variables such as arrays, various stacks and registers, and system setting values.

  Among such storage means, the related word storage means 25 stores a plurality of simultaneously used words that are input by being separated by a space when performing an AND search with a search query as related words. Means. In the example of FIG. 2, a group of related words (eg, “bred”, “picture book”, etc.) is associated with “snake” for each reference word (eg, “snake”), but the actual format is arbitrary. Thus, a graph index or the like indicating the distance between multiple words may be used instead of each reference word.

  As illustrated in FIG. 3, the correspondence dictionary storage unit 75 associates each type of attribute word that may be extracted for the search query with the corresponding action candidate pattern, search character string pattern, and the like. , Is a means for storing. The index storage means 85 is means for storing index data for web search according to the prior art. Each means other than the storage means is a processing means for realizing and executing the following information processing functions and operations.

[2. Action)
FIG. 4 is a flowchart showing the operation of the present embodiment configured as described above.
[2-1. (Attribute extraction)
First, when the search query acquisition unit 10 acquires a word (keyword) input in a search window of a so-called web search engine as a search query (step S1), the attribute candidate generation unit 20 is based on the acquired search query. Then, an attribute candidate word that is a word candidate representing an attribute of the search intention related to the search query is generated (step S2). The attribute candidate words are, for example, related words that are easily AND-searched with the words of the search query. Here, the attribute candidate generating means 20 is used as the input search query and related words from the related word storage means 25 (FIG. 2). The associated words are extracted as attribute candidate words (for example, FIG. 5 (1)).

  Subsequently, the pattern processing unit 30 extracts the attribute of the search intention by performing word processing with a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms on the generated attribute candidate word. (Step S3). More specifically, the pattern processing means 30 removes the proper noun from the attribute candidate word as the word processing by the proper noun removal means 31 which is difficult to become an attribute (for example, “Miyakojima” in FIG. 5 (2)), Further, after removing the noun adjectives by the noun adjective removing means 32 which is difficult to become an attribute (for example, “giant” in FIG. 5 (3)), the synonym processing means 33 unifies the synonyms for the remaining attribute candidate words. (For example, in FIG. 5 (4), “how to keep”, “dream fortune-telling”, and “extermination” are removed), and attribute candidate words remaining as a result of the removal and unification are set as attribute words representing the attribute of the search intention. The synonyms here are not limited to complete synonyms, but include words having a predetermined relationship as similar themes.

[2-2. Action candidate generation and output)
Subsequently, the action candidate generating means 40 generates action candidates corresponding to the attribute words extracted as attributes (referred to as “action candidates”) (step S4), and the generated action candidates are output as action candidates. The means 50 outputs (step S5). The remaining attribute words as shown in FIG. 5 (4) are preferably used as action candidates in the order corresponding to the rearrangement result after being rearranged (reranking) by the reranking means 38. .

  Sorting criteria are free, but for example, based on the word of the search query, it can be sorted by a statistical index (correlation coefficient, mutual information amount, etc.) based on the number of appearances and co-occurrence of each attribute word. In addition, for example, for each word of the same query, initially, each action candidate corresponding to the attribute word is actually displayed randomly to various users, and as a result, a click rate (CTR: click through. It is conceivable to output action candidates corresponding to each in the order of priority selected from attribute words having a high rate).

For example, the remaining attribute words “breeding, picture book, dream, pet, type, chick, extermination, princess, image ...” in the example of FIG. 5 (4) based on the search query “snake” are arranged as described above. As a result of the change, consider the case where the order is "type, extermination, dream, illustration, image, breeding ...". In this case, according to the type of each attribute word, the query word and the attribute word are applied to the action candidate pattern stored in the correspondence dictionary storage means 75 illustrated in FIG.

"I want to know about snake types"
"I want to know about snake extermination"
"I want to know about snake dreams"
"I want a snake illustration"

The action candidates generated as described above are output as web pages as illustrated in FIG.

  In the example of FIG. 6, the user inputs “snake” as the search query X1 in the input field F of the search engine, and the search button B1 is operated with the mouse pointer P. As a result, several action candidate CAs are displayed. Show. In this figure, the underline attached to each action candidate is a hyperlink of a web page for transitioning to a search result page when the search is further advanced by specifying the action. In the figure, an appropriate explanatory text regarding the related content is displayed along with each action candidate, but such an explanatory text is not essential.

  Also, if there are multiple (for example, two) action candidate patterns, such as “I want to see a snake's dream” and “I want to know a snake's dream”, whether to output both or one of them, and the selection criteria Is also free. For example, both candidates may be randomly output to various users, and may be unified into candidates that have been clicked a lot as the entire apparatus. For example, if “... I want to know” is selected more frequently than “...”, the output is unified into the latter.

[2-3. (Reception and search of actions)
When the action accepting unit 60 accepts designation of one of the action candidates output as described above (step S6), the search generating unit 70 retrieves the search character string corresponding to the designated action. (Step S7), the search means 80 executes a search using the generated search character string (step S8), and the result output means 90 outputs the result of the executed search (step S8). S9).

For example, let us consider a case where “I want to know about snake removal” in the action candidate CA of FIG. 6 is designated by a hyperlink click selection operation or the like. The attribute word “disinfecting” corresponding to this action candidate is “service name” in the type of attribute word shown in FIG. 3, and the pattern of the corresponding search character string is

[Web search]: {query word} and ({attribute word} or {synonym}) and “I want to know”

It is.

In addition to the query word “snake”, attribute word “disinfect”, and the synonym “defeat” obtained from the synonym or synonym data removed earlier, the search string is

[Web Search]: Snake and (exterminate or extermination) and want to know

The search result based on this is as shown in FIG. In this search result screen example, the display in the input field F is the character string “snake and (disinfection or extermination) and want to know” used for the search, and each link L to the corresponding web page is displayed. Yes.

  Note that the search condition corresponding to the action “I want to know” is also included in the search expression, but “Snake and (disinfection or extermination)” may yield similar results to some extent. Even if there is a phrase “I want to know” indicating the action in the action candidate), it is not essential to put the action in the search expression.

  Here, the search character string corresponding to the action is not limited to a combination of search words, for example, and may have a function of switching the type of search engine or search service such as an API URL. For example, as illustrated in FIG. 3, [Web search] [shopping search] or the like (actually, an API URL corresponding to each search type) is switched as a search character string pattern according to the type of attribute word. If a character string for this is set, a search result highly compatible with the search intention can be provided to the user.

[2-4. (Additional keywords and image output)
Further, the action candidate output unit 50 may perform a search using a search character string corresponding to the action candidate and output an additional keyword based on the search result together with the action candidate. For example, in the screen example of FIG. 8, hyperlinks such as “carnation” and “stall” are added as additional keywords AA in relation to the action A corresponding to the attribute word “present” extracted for the search query “Mother's Day”. It is displayed.

  As such additional keywords, for example, if you click on “Carnation” searched for “Mother's Day Gift Purchase”, you can immediately switch to the search result page for “Mother's Day Gift Purchase” and “Carnation” It is convenient to do so. For example, these additional keywords are extracted from a web page of a search result or a web page indicated by a URL included in the search result by applying a technique such as a known TF / IDF.

  The action candidate output means 50 may perform a search using a search character string corresponding to the action candidate, and output an image based on the search result together with the action candidate. For example, in the screen example of FIG. 9, “carnation” and “stall” which are additional keywords AA according to FIG. 8 are associated with the action A corresponding to the attribute word “present” extracted for the search query “Mother's Day”. In addition to the above, an image G of carnation or stall is displayed. Similar to the additional keyword, when such an image G is clicked, for example, a similar image search or the like may be performed in a shopping site.

  As an image based on the search result as described above, for example, a feature word is extracted from a search result web page or a web page indicated by a URL included in the search result by applying a known technique such as TF / IDF. The result of image search using the feature word can be used.

  The action candidate output means 50 performs a search using a search character string corresponding to the action candidate before suggesting (outputting action candidates) to the user in advance, and the search results are less than a predetermined number (for example, 0 cases). Note that “less than a predetermined number” includes “below the predetermined number”) action candidates may be excluded from display targets.

[3. effect〕
As described above, in this embodiment, based on a search query, an attribute candidate word that is a word candidate representing a search intention attribute is extracted, and a predetermined type of word is removed from the attribute candidate word. (For example, FIG. 5), extracting the attributes of the search intention in the form of remaining candidate words and combinations thereof, and presenting the candidates for actions corresponding to the attributes (for example, FIG. 6), The extracted search intention attributes can be reflected in the search, such as switching the search type and conditions according to the selection of candidates, so that search results that match the user's search intention can be provided without excessive user load such as long text input. be able to.

  In particular, in the present embodiment, a search for a user is performed by presenting a candidate for an action according to a search intention attribute extracted from a search query to the user and generating an appropriate search character string according to the selection from the candidate. Since the search result suitable for the intention can be displayed, it is possible to more effectively provide the search result that matches the search intention of the user without excessive user load.

  Further, in this embodiment, related words that are often combined based on the search query are attribute candidate words, and proper nouns, nounative adjectives, and synonyms are excluded (for example, FIG. 5). The attribute of the search intention can be efficiently extracted by a simple process to provide a search result that matches the user's search intention.

  Furthermore, in the present embodiment, a search is performed in advance using a search character string for each action candidate, and additional keywords such as more specific options are output together with the candidate based on the search result (for example, FIG. 8). Since it becomes easy to narrow down information to be acquired, usability of search is improved.

  In addition, in the present embodiment, a search is performed in advance using a search character string for each action candidate, and an image such as a product photo for each product is output together with the candidate based on the search result (for example, FIG. 9). The contents of the information provided when “” is designated can be easily understood from the image at a glance, and the usability of the search is improved.

  In addition, in this embodiment, when there is no search result for the action, the action is excluded from the display target, so that a useless operation of 0 search results for the selected candidate can be avoided, and the usability of the search is improved. To do.

[4. Other embodiments]
In addition, said each embodiment is only an illustration, and this invention includes what is illustrated below and other embodiment other than that. For example, in the above embodiment, the generation and output of action candidates corresponding to the extracted attributes are indispensable. However, the effects of the present invention can be obtained even if these are not essential and only the attribute extraction is performed.

  In this case, the search device according to the present invention includes a search query acquisition unit that acquires a search query, and an attribute candidate word that is a candidate for a word that represents an attribute of a search intention related to the search query based on the acquired search query. The attribute of the search intention is generated by performing word processing with a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms on the generated attribute candidate word and the generated attribute candidate generation unit. Pattern processing means for extracting. For example, FIG. 10 is an example of a screen that displays only the attribute display Z of the search intention such as the “type” of the snake and the “disinfection” of the snake as options.

  In this way, based on the search query, an attribute candidate word that is a candidate word representing the attribute of the search intention is extracted, and word processing such as removing a predetermined type of word is performed on the attribute candidate word, By extracting the search intention attributes in the form of remaining candidate words and their combinations, and presenting the search query and attribute combinations according to those attributes to the user, the search type and conditions are switched according to the combination selection Thus, the extracted search intention attribute can be reflected in the search, so that a search result suitable for the user's search intention can be provided without excessive user load such as long text input.

  In addition, each element such as the unit illustrated in FIG. 1 or the like is not limited to the computer control unit, and may be realized by another information processing mechanism such as an electronic circuit based on wired logic or the like. Further, each configuration diagram, data diagram, flowchart diagram, and the like are merely examples, and the presence / absence of each element, its order, specific contents, and the like can be changed as appropriate. For example, the apparatus of the present invention may be realized by using a plurality of apparatuses such as servers, and a configuration in which each storage unit is realized by a separate and independent server apparatus or system is also common. Depending on the function, the configuration can be flexibly changed, for example, by calling an external platform or the like with an API (application program interface) or network computing (so-called cloud or the like).

DESCRIPTION OF SYMBOLS 1 Search apparatus 6 Arithmetic control part 7 Storage apparatus 8 Communication means 10 Search query acquisition means 20 Attribute candidate generation means 25 Related word storage means 30 Pattern processing means 31 Proper noun removal means 32 Noun adjective removal means 33 Synonym processing means 38 Ranking means 40 Action candidate generation means 50 Action candidate output means 60 Action reception means 70 Search generation means 75 Corresponding dictionary storage means 80 Search means 90 Result output means CA Action candidate A Action B1 Search button F Input field G Image L Link N Communication network P Mouse pointer T Terminal X1 Search query

Claims (6)

A search query acquisition means for acquiring a search query from a terminal ;
Based on the acquired search query, attribute candidate generation means for generating an attribute candidate word that is a word candidate representing an attribute relating to the search query;
Pattern processing means for extracting the attribute by performing word processing with a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms for the generated attribute candidate word;
Action candidate generation means for generating an action candidate corresponding to the extracted attribute;
Action candidate output means for performing a search using a search character string corresponding to the generated action candidate and outputting an additional keyword or an image based on the search result to the terminal together with the generated action candidate ,
When any one of the action candidates is designated, a search is performed using a search character string corresponding to the designated action, and when the additional keyword or image is designated, A search means for executing a search using the specified additional keyword or image;
And a result output means for outputting the result of the executed search to the terminal . The search means, when the additional keyword is specified, performs a search using the specified additional keyword and a search character string corresponding to the action candidate. The search device according to 1. A related word storage means for storing a plurality of words used simultaneously in the search query in association with each other as related words;
The attribute candidate generation unit extracts, as the attribute candidate word, a word associated with the input search query and the related word from the related word storage unit,
The pattern processing means includes
As the word processing, proper noun removal means for removing proper nouns from the attribute candidate words,
As the word processing, nounative adjective removal means for removing nounative adjectives from the attribute candidate words,
Synonym processing means for unifying synonyms for the attribute candidate words as the word processing;
The search apparatus according to claim 1 or 2, wherein the attribute candidate words remaining as a result of the removal and unification are used as the attribute. The action candidate output means performs a search using the search character string corresponding to the action candidates, and excludes action candidates whose search results are less than a predetermined number from display targets. The search device according to any one of 1 to 3 . A search query acquisition step in which a computer acquires a search query from a terminal ;
An attribute candidate generation step in which the computer generates attribute candidate words that are word candidates representing attributes related to the search query based on the acquired search query;
A pattern processing step in which the computer extracts the attribute by performing a word process with a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms for the generated attribute candidate word;
An action candidate generating step in which a computer generates an action candidate corresponding to the extracted attribute;
Action candidate that computer performs search using search character string corresponding to generated action candidate and outputs additional keyword or image based on search result to said terminal together with generated action candidate An output step;
When any one of the action candidates is designated, the computer executes a search using a search character string corresponding to the designated action, and the additional keyword or image is designated. A search step for performing a search using the specified additional keyword or image,
And a result output step in which a computer outputs a result of the executed search to the terminal . By controlling the computer
Get a search query from your device ,
Based on the obtained search query, an attribute candidate word that is a word candidate representing an attribute related to the search query is generated,
For the generated attribute candidate word, the attribute is extracted by performing word processing with a predetermined pattern including at least one of removal of a predetermined type of word and unification of synonyms,
Generate action candidates corresponding to the extracted attributes,
Performing a search using a search character string corresponding to the generated action candidate, causing an additional keyword or an image based on the search result to be output to the terminal together with the generated action candidate ,
When any one of the action candidates is designated, a search is performed using a search character string corresponding to the designated action, and when the additional keyword or image is designated, Perform a search using the specified additional keyword or image,
A search program for causing the terminal to output a result of the executed search.

Images