WO2011013229A1

WO2011013229A1 - Behavior recommendation device

Info

Publication number: WO2011013229A1
Application number: PCT/JP2009/063577
Authority: WO
Inventors: 昌之岡本; 貴之飯田; 匡晃菊池; 奈夕子渡辺
Original assignee: 株式会社東芝
Priority date: 2009-07-30
Filing date: 2009-07-30
Publication date: 2011-02-03

Abstract

A behavior recommendation device is provided with an appearance frequency identification section (206) which identifies the weights of behavior expressions of an unknown word from a document containing the unknown word; a characteristic expression storage section (210) which stores an existent characteristic expression, the behavior expressions of the characteristic expression, and the appearance frequencies of the behavior expressions in association; a meaning estimation section (212) which estimates the meaning of the unknown word according to the degree of similarity between the appearance frequencies of the behavior expressions of the unknown word and those of the behavior expressions of the characteristic expressions; a search method storage section (214) which stores the behavior expressions and search methods in association; a search method selection section (216) which identifies the behavior expression associated with the meaning estimated in the meaning estimation section (212), and selects the search method associated with the behavior expression in the search method storage section (214); and an output section which outputs the result of search.

Description

Action recommendation device

The present invention relates to an action recommendation that provides appropriate information to a user based on an unknown word.

Conventionally, for words appearing in text information (hereinafter referred to as documents) such as web pages and manuscripts, there is a demand to search related information according to the meaning and easily recommend information appropriate for the user. As this type of device, a technique for estimating the meaning of a keyword based on the similarity between statistical information of nearby words and the thesaurus concept has been proposed (see, for example, Patent Document 1).

Japanese Patent Laid-Open No. 11-212975

However, there are cases where the word appearing in the text information is a new word, that is, an unknown word, and it is difficult to determine what information should be recommended to the user for the unknown word. It was.

The present invention has been made in view of the above, and an object thereof is to provide an action recommendation device capable of providing appropriate information to a user based on an unknown word.

In order to solve the above-described problems and achieve the object, the present invention is an action recommendation device, which extracts an unknown word that matches a predetermined condition from content including a document, An appearance frequency specifying unit that specifies the weight of one or more action expressions connected to the unknown word from the document including the unknown word, an existing specific expression, and 1 or 2 connected to the specific expression The specific expression storage unit that associates and stores the behavior expression and the weight of the behavior expression in the document including the behavior expression, and the one or more of the behavior expressions that are connected to the unknown word. A semantic estimation unit that estimates the meaning of the unknown word based on a similarity between a weight and the weight of one or more of the behavioral expressions that are connected to the specific expression; the behavioral expression; and a search method With correspondence In the search method storage unit and the specific expression storage unit, the behavior expression associated with the meaning estimated in the meaning estimation unit is specified, and the search method storage unit stores the behavior expression A search method selection unit for selecting the search method associated with the search method; a search unit for searching for the unknown word by the search method selected by the search method selection unit; and a search result output by the search unit And an output unit.

According to the present invention, it is possible to provide appropriate information to the user based on the unknown word.

The block diagram which shows the structure of the action recommendation system 1. FIG. The figure which shows the data structure of the unknown word storage part 208. FIG. The figure which shows the data structure of the specific expression storage part 210. FIG. The figure which shows the data structure of the search method storage part 214. FIG. The flowchart which shows an action recommendation process. The figure which shows the search table 220. FIG. The figure which shows an action recommendation screen. The block diagram which shows the structure of the action recommendation system 2. FIG. The figure which shows the data structure of the specific expression storage part 238.

Hereinafter, with reference to the attached drawings, the best embodiment of the behavior recommendation device according to the present invention will be described in detail.

FIG. 1 is a block diagram showing the configuration of the behavior recommendation system 1. The behavior recommendation system 1 extracts an unknown word from the mobile terminal 100 used by the user and text information in the content referred to by the mobile terminal 100, and performs processing for providing the user with a service suitable for the unknown word. And an action recommendation device 200. Here, the unknown word is a word whose meaning is not registered in the behavior recommendation device 200 or the like. Specifically, there are combinations of plural nouns, words in katakana notation, and the like.

The mobile terminal 100 includes a communication unit 102, a display unit 104, a content extraction unit 106, and a selection unit 108. The communication unit 102 transmits and receives various information via the Internet. For example, the communication unit 102 transmits the position information of the mobile terminal 100 obtained by GPS (Global Positioning System) to the behavior recommendation device 200. The display unit 104 displays a web page acquired by the communication unit 102 via the Internet. The web page includes still images, moving images, and the like in addition to text information.

The content extraction unit 106 extracts the display content displayed on the display unit 104. The selection unit 108 selects information to be displayed on the display unit 104 in accordance with an input from the user. Note that the information selected by the selection unit 108 includes information provided by the behavior recommendation device 200.

The behavior recommendation device 200 includes a communication unit 202, an unknown word extraction unit 204, an appearance frequency specifying unit 206, an unknown word storage unit 208, a specific expression storage unit 210, a meaning estimation unit 212, and a search method storage unit 214. And a search method selection unit 216 and a search unit 218.

The communication unit 202 transmits and receives various information via the Internet. The communication unit 202 receives, for example, a document that the user is browsing on the mobile terminal 100, that is, text information or display content of a web page displayed on the display unit 104. The unknown word extraction unit 204 identifies a word that matches a predetermined notation condition as an unknown word from the text information and display content received by the communication unit 202 from the mobile terminal 100. Examples of the display condition include parenthesis notation, meaning that the meaning of a word is not registered in the behavior recommendation device 200, and the like.

The appearance frequency specifying unit 206 weights one or more action expressions having a linguistic connection relationship or a semantic connection relationship with an unknown word extracted by the unknown word extraction unit 204 for a predetermined document. The action frequency is specified as Specifically, the appearance frequency specifying unit 206 specifies the appearance frequency of one or more action expressions that co-occur with an unknown word as an object. Here, the action expression is a verb. Existing morphological analysis and syntax analysis techniques can be used to extract co-occurring behavioral expressions. For example, if an unknown word “vejimite” contains two sentences “eating vegemite” and “want to eat vegemite” in the target document, the action expression “eating” The appearance frequency is specified as 2 times. In this embodiment, an unknown word is acquired using an action expression with the unknown word as the object, but a co-occurrence relationship not limited to the object or a connection expression other than the object may be used. . Moreover, although the appearance frequency itself is used as the weight of the action expression, a value obtained by normalizing the appearance frequency with the appearance frequency of all the action expressions may be used.

Note that the document targeted when the appearance frequency specifying unit 206 specifies the appearance frequency of an unknown word is a web page arbitrarily acquired via a network. As another example, the behavior recommendation device 200 or another device has a document storage unit that stores a plurality of documents, and the appearance frequency specifying unit 206 targets documents stored in the document storage unit. It is good also as specifying the appearance frequency of an unknown word. Thus, the appearance frequency specifying unit 206 only needs to be able to specify the appearance frequency of unknown words in an arbitrarily selected document, and the target document is not particularly limited.

The appearance frequency specifying unit 206 stores the specified appearance frequency in the unknown word storage unit 208 together with the unknown word. As illustrated in FIG. 2, the unknown word storage unit 208 stores, for each unknown word, the co-occurring behavioral expression and the appearance frequency of the unknown word in association with each other. In this way, the appearance frequency specifying unit 206 makes it possible to identify unknown words for one or more action expressions such as “vejimite” 20 times for the action expression “eat” and 13 times for “paint”. Distribution of appearance frequency is obtained.

As shown in FIG. 3, the specific expression storage unit 210 stores a noun as an existing specific expression, one or more action expressions co-occurring in the specific expression, and the appearance frequency of the specific expression in association with each other. ing. Each unique expression belongs to a semantic class, and the semantic class is hierarchically structured. For example, the specific expression “butter” belongs to the semantic class “food”. Further, the semantic class “food” belongs to the semantic class “food”.

For example, with respect to the unique expression “butter”, the specific expression storage unit 210 has an appearance frequency of 477 times that co-occurs with “use”, an appearance frequency of 354 times that co-occurs with “put”, and an appearance frequency of 309 times that co-occurs with “paint”. As described above, a plurality of behavioral expressions co-occurring with a specific expression and the appearance frequency of each specific expression are stored in association with each other. That is, the specific expression storage unit 210 stores the specific expression “butter” and the distribution of the appearance frequencies for each of the plurality of behavior expressions that co-occur with the specific expression “butter”. The same applies to the other proper expressions “jam” and “hamburg”.

Furthermore, the behavioral expression and the appearance frequency of each semantic class are the sum of the appearance frequencies for a plurality of specific expressions belonging to each semantic class. For example, the frequency of appearance of the action expression “use” for the semantic class “food” is 477 times and 69 times of appearance of the action expressions “use” of the specific expressions “butter” and “jam” belonging to the meaning class “food”. Is the sum of Similarly, the action expression and the appearance frequency of the semantic class in the upper hierarchy are the sum of the appearance frequencies for the specific expressions belonging to the plurality of semantic classes belonging to the semantic class in the upper hierarchy. For example, the frequency of appearance of the action expression “use” in the semantic class “food” is the specific expressions “butter” and “jam” belonging to the lower semantic classes “food” and “dish” included in the semantic class “food”. ”,“ Hamburger ”and“ pasta ”are the total appearance frequencies of the action expression“ use ”. In this embodiment, the sum of the appearance frequencies of the behavior expressions of the lower semantic classes is used as the weight of the behavior expressions of the higher semantic classes. However, the weights of the behavior expressions of the respective semantic classes are not normalized. It is also possible to have a plurality of operations in between, such as the sum of the normalized values and the normalized value.

The meaning estimation unit 212 distributes the appearance frequency of one or more action expressions for the unknown word obtained by the appearance frequency specifying unit 206 and one or more for the specific expressions stored in the specific expression storage unit 210. The appearance frequency distribution is compared, and the proper expression similar to the unknown word or the semantic class to which the unknown word belongs is estimated as the meaning of the unknown word.

Note that the inner product of vectors is used for the evaluation of similarity. That is, the appearance frequency distribution for the behavioral expression is regarded as a word vector, and it is determined that the larger the inner product, the more similar. Specifically, the product of the appearance frequencies of the (identical) action expressions common to the unknown word and the specific expression to be compared is calculated. Then, the sum of all products obtained for each of the behavioral expressions common to the unknown word and the specific expression to be compared is calculated as the similarity. In this embodiment, a vector inner product is used as the similarity, but the Euclidean distance is not used even if a similarity calculation method other than the inner product itself is used, such as assigning a weight to a specific action expression. The similarity calculation method may be used.

Here, a case where the unknown word “vejimite” is compared with the specific expression “butter” stored in the specific expression storage unit 210 will be described. When the unknown word “vejimite” has an appearance frequency as shown in FIG. 2 and the unique expression “butter” has an appearance frequency as shown in FIG. “20 × 223” is calculated as the product of the appearance frequencies. Furthermore, “13 × 309” is calculated as the product of the appearance frequencies for “paint”. The sum of products for all common behavior expressions calculated in this way is obtained as the similarity.

As shown in FIG. 4, the search method storage unit 214 stores an action expression and a search method for an unknown word in association with each other. The search method is a method for searching for an action recommended for the user of the mobile terminal 100. For example, if the unknown word co-occurs with the action expression “buy”, the corresponding search method is, for example, “search for a product site that uses an unknown word as a product and present the URL of the site”. is there. Thereby, the user can recommend an action of browsing the product site.

The search method selection unit 216 refers to the specific expression storage unit 210 and identifies an action expression corresponding to the meaning estimated by the meaning estimation unit 212. The search method selection unit 216 further selects a search method associated with the identified action expression in the search method storage unit 214.

The search unit 218 performs a search for an unknown word to be processed by the search method selected by the search method selection unit 216, and transmits the search result to the mobile terminal 100 via the communication unit 202. That is, the communication unit 202 corresponds to an output unit that outputs search results.

FIG. 5 is a flowchart showing a behavior recommendation process by the behavior recommendation device 200. When the communication unit 202 of the behavior recommendation device 200 receives browsing information including text information such as a web page being browsed from the mobile terminal 100, the unknown word extraction unit 204 uses the text information included in the browsing information received by the communication unit 202. An unknown word is extracted (step S100). When the unknown word extracted by the unknown word extraction unit 204 is not stored in the unknown word storage unit 208 (No in step S102), the appearance frequency specifying unit 206 targets the web page as a predetermined document. The appearance frequency of the action expression co-occurring with the unknown word is specified (step S104). Next, the appearance frequency specifying unit 206 stores the specified behavioral expression and the appearance frequency in the unknown word storage unit 208 in association with the unknown word (step S106).

Next, the semantic estimation unit 212 determines the distribution of the appearance frequency of one or more action expressions for the unknown word and the appearance frequency of one or more action expressions for the specific expression stored in the specific expression storage unit 210. The similarity of the distribution is calculated, and the meaning is estimated based on the similarity (step S120). Specifically, the specific expression having the highest similarity is estimated as the meaning of the unknown word. For example, if the similarity between the unknown word “vejimite” and the specific expression “butter” stored in the specific expression storage unit 210 is highest, it is determined that the unknown word “vejimite” is a word close to “butter”. .

When the degree of similarity with the most similar specific expression is lower than a predetermined threshold value, that is, when it is not similar to any specific expression stored in the specific expression storage unit 210, the semantic estimation unit 212 displays the specific expression. The similarity between the semantic class to which the word belongs and the unknown word is calculated, and the meaning of the semantic class having the highest similarity with the unknown word is estimated as the meaning of the unknown word. For example, when the degree of similarity with any specific expression such as “butter” or “jam” is low, the degree of similarity between the semantic class “food” and the unknown word is calculated. Thereby, although it cannot be estimated that the meaning is similar to a specific food such as “butter” or “jam”, it can be estimated that it is “food”.

Furthermore, when the similarity between each semantic class in the same hierarchy as the semantic class “food” and the unknown word is lower than the threshold, the similarity between the semantic class and the unknown word in the higher hierarchy is calculated, and the unknown word and The meaning of the semantic class with the highest similarity is estimated as the meaning of the unknown word. Thus, when the degree of similarity is equal to or less than the threshold value, appropriate meaning estimation can be performed by increasing the abstraction level and estimating the meaning.

Next, the search method selection unit 216 specifies an action expression associated with the meaning estimated by the meaning estimation unit 212 in the specific expression storage unit 210. Then, the search method storage unit 214 selects a search method associated with the action expression (step S122). Next, the search unit 218 searches for an object using the search method selected by the search method selection unit 216, and transmits the search result to the mobile terminal 100 (step S124). That is, behavior recommendation is performed to the user. This completes the action recommendation process.

Suppose, for example, that the unknown word “vejimite” has the highest similarity with the proper expression “butter” and a similarity equal to or higher than a threshold is obtained. In this case, in step S120, the meaning estimation unit 212 estimates that the unknown word “vejimite” is a word having a meaning close to “butter”. In step S <b> 122, the search method selection unit 216 refers to the specific expression storage unit 210 and identifies an action expression associated with “butter” estimated to have a meaning similar to the unknown word “vegetite”. In the example shown in FIG. 3, a plurality of behavioral expressions such as “buy” and “eat” are specified. The search method selection unit 216 further refers to the search method storage unit 214 and specifies a search method associated with these action expressions. In the example shown in FIG. 4, since the search methods are associated with “buy” and “eat” among the behavioral expressions associated with “butter”, these search methods are selected. .

The search method selection unit 216 may select a plurality of search methods or one search method. For example, the search method storage unit 214 may select all the search methods associated with the target behavioral expression. As another example, it is possible to preferentially select a search method for a behavioral expression having a higher appearance frequency among the behavioral expressions specified in the specific expression storage unit 210.

When the search method associated with “Buy” and “Eat” is selected from the search method storage unit 214 shown in FIG. 4, in step S124, the search unit 218 follows the search method and the unknown word “vejimite”. Is used as a query to search a merchandise sales site, and information indicating the URL is transmitted to the mobile terminal 100. In the mobile terminal 100, this URL is displayed on the display unit 104. When the selection unit 108 selects a URL by an input from the user, a web page corresponding to the URL is displayed on the display unit 104. Thereby, it is possible to recommend an action of purchasing “Vegemite” on the Internet to the user.

Further, according to the search method, the search unit 218 searches for a “Vegemite” merchandise sales site, and acquires a sales store and its position. Then, when a sales store exists within a radius of 100 m from the current position of the mobile terminal 100, information related to the sales store is displayed on the display unit 104 of the mobile terminal 100. This makes it possible to recommend the user the action of purchasing “Vegemite” at the store.

Similarly, according to the search method, the search unit 218 searches for a restaurant site where “vejimite” can be eaten, and acquires the restaurant name and its position. Then, when a restaurant exists within a radius of 100 m from the current position of the mobile terminal 100, information about the restaurant is displayed on the display unit 104 of the mobile terminal 100. As a result, it is possible to recommend the user an action of eating “vejimite” at a restaurant.

Specifically, the search unit 218 creates a search table 220 as shown in FIG. 6 based on the search method selected by the search method selection unit 216 and the unknown word. In the search table 220, latitude and longitude as location information of a store, recommended contents, and recommended conditions are associated with unknown words and behavioral expressions. The search unit 218 periodically compares the current position information received from the mobile terminal 100 via the communication unit 202, the position information associated with the unknown word in the search table 220, and the recommendation condition, and recommends If the condition is met, the recommended content is transmitted to the mobile terminal 100 via the communication unit 202. That is, action recommendation is performed when the recommendation condition is met.

For example, an action recommendation screen as shown in FIG. 7 is displayed on the display unit 104 of the mobile terminal 100. As described above, since the map including the current position and the position of the store or restaurant and the information indicating the name of the store or restaurant are displayed, the user can obtain action recommendation information related to the unknown word “vejimite”. it can.

As described above, the behavior recommendation system 1, when the unknown word is extracted, the appearance frequency of the behavior expression co-occurring with the specific expression stored in the specific expression storage unit 210 and the appearance of the behavior expression co-occurring with the unknown word. It is possible to specify a specific expression that is similar in frequency, and estimate the meaning of the specific expression to be close to the meaning of the unknown word. Furthermore, behavior recommendation suitable for the estimated meaning can be performed to the user. In other words, even when the meaning of an unknown word cannot be directly determined by language processing, it is possible to estimate the meaning based on the usage of the unknown word and perform appropriate action recommendation.

On the other hand, when the unknown word extracted by the unknown word extraction unit 204 is already stored in the unknown word storage unit 208 in step S102 of FIG. 5 (step S102, Yes), the appearance frequency specifying unit 206 is the target. Instead of specifying the appearance frequency of the unknown word action expression in the document, the unknown word storage unit 208 specifies the appearance frequency of the action expression associated with the unknown word (step S110). As described above, when an unknown word is already stored in the unknown word storage unit 208, the processing efficiency can be improved by using the information stored in the unknown word storage unit 208.

As a first modification of the present embodiment, the mobile terminal 100 and the behavior recommendation device 200 may be provided integrally. That is, the mobile terminal 100 may extract unknown words, estimate the meaning of unknown words, and perform processing related to behavior recommendation suitable for the meaning.

As a second modification example, the appearance frequency specifying unit 206 specifies the appearance frequency of the action expression using a web page or the like viewed by the user of the mobile terminal 100 as a target document, and the unknown word storage unit 208 For each user identification information for identifying a user of the terminal 100, an action expression and an appearance frequency for an unknown word may be stored. Thereby, the action expression and appearance frequency which co-occur on an unknown word within the range browsed by the user can be obtained. Therefore, it is possible to identify an appropriate action expression and make an appropriate action recommendation.

As shown in FIG. 8, the behavior recommendation device 230 of the behavior recommendation system 2 according to the second exemplary embodiment further includes a situation information specifying unit 232. The communication unit 202 of the behavior recommendation device 230 receives status information indicating that the user is on a train or the like in addition to the position information from the mobile terminal 100. Note that the mobile terminal 100 specifies whether or not the user is on the train based on the position information of the mobile terminal 100 obtained by GPS and the detection result by the acceleration sensor provided in the mobile terminal 100, and gets on the train. Generate status information indicating that it is in progress. The communication unit 202 also receives, from the mobile terminal 100, browsing time when the user browsed the web page as status information.

The status information specifying unit 232 specifies the status information received by the communication unit 202 from the mobile terminal 100. The situation information specifying unit 232 may further specify information held by itself such as the current time as the situation information.

The appearance frequency specifying unit 234 specifies the appearance frequency of the action expression using the situation information specified by the situation information specifying unit 232 as a query together with the unknown word. For example, status information such as “morning” in the morning and “train” when in the train is added to the query. This makes it possible to extract the appearance frequency of behavioral expressions related to the situation information, for example, “Vegemite” is often eaten in the morning from expressions such as “I ate vegemite in the morning”.

The appearance frequency specifying unit 234 further stores the appearance frequency of the action expression in the unknown word storage unit 236 in association with the situation information. That is, in the unknown word storage unit 236, as shown in FIG. 9, the behavioral expressions “use”, “put”, and “paint” for the unknown word “vejimite” have appearance frequencies “morning”, “daytime”, “night”. "Is stored in association with each status information. As a result, as shown in FIG. 9, the frequency of appearance of action expressions corresponding to the situation information is obtained, such that “paint” often co-occurs in the morning and “buy” often co-occurs in the daytime. be able to.

Also in the unique expression storage unit 238, as in the unknown word storage unit 236, the appearance frequency of the action expression for the specific expression is stored in association with each situation information.

The meaning estimation unit 240 generates the appearance frequency of the action expression for the unique expression associated with the situation information that matches the current situation information, and the appearance of the action expression for the unique expression of the unknown word obtained by the appearance frequency specifying unit 234. Based on the frequency similarity, the meaning of the unknown word is estimated.

The search method selection unit 242 specifies an action expression associated with the current situation information in the specific expression storage unit 238. Then, the search method storage unit 214 selects the search method associated with the action expression.

Accordingly, for example, in the morning, an action recommendation related to “painting” such as presentation of a URL of a cooking recipe including “vejimite” associated with “painting” that often co-occurs with “morning” is provided. It can be carried out. In the daytime, it is possible to make an action recommendation such as presenting the URL of the store where “Vegemite” is associated with “Buy” that often co-occurs with “Day”. The remaining configuration and processing of the behavior recommendation system 2 according to the second embodiment are the same as the configuration and processing of the behavior recommendation system 1 according to the first embodiment.

The behavior recommendation device according to the present embodiment includes a control device such as a CPU, a storage device such as a ROM (Read Only Memory) and a RAM, an external storage device such as an HDD and a CD drive device, and a display device such as a display device. It has an input device such as a keyboard and a mouse, and has a hardware configuration using a normal computer.

The behavior recommendation program executed by the portable terminal and the behavior recommendation device of the present embodiment is a file in an installable or executable format, and is a CD-ROM, flexible disk (FD), CD-R, DVD (Digital Versatile Disk). And the like recorded on a computer-readable recording medium.

Further, the behavior recommendation program executed by the mobile terminal and the behavior recommendation device of the present embodiment may be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network. good. Moreover, you may comprise so that the action recommendation program performed with the action recommendation apparatus of this embodiment may be provided or distributed via networks, such as the internet. Moreover, you may comprise so that the action recommendation program of this embodiment may be previously incorporated in ROM etc. and provided.

The behavior recommendation program executed by the behavior recommendation device according to the present embodiment is a module including the above-described units (communication unit, unknown word extraction unit, appearance frequency identification unit, meaning estimation unit, search method selection unit, search unit) and the like. As actual hardware, a CPU (processor) reads out and executes an action recommendation program from the storage medium, so that each unit is loaded on the main storage device, and a communication unit, an unknown word extraction unit, An appearance frequency identification unit, a meaning estimation unit, a search method selection unit, and a search unit are generated on the main storage device.

Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

DESCRIPTION OF SYMBOLS 1 Behavior recommendation system 100 Portable terminal 102 Communication part 104 Display part 106 Content extraction part 108 Selection part 200 Behavior recommendation apparatus 202 Communication part 204 Unknown word extraction part 206 Appearance frequency specific part 208 Unknown word storage part 210 Specific expression storage part 212 Meaning estimation Unit 214 Search method storage unit 216 Search method selection unit 218 Search unit 220 Search table

Claims

An unknown word extraction unit that extracts an unknown word that matches a predetermined condition from content including a document;
An appearance frequency specifying unit that specifies weights of one or more behavioral expressions connected to the unknown word from a document including the unknown word;
A specific expression storage unit that associates and stores an existing specific expression, one or more of the action expressions connected to the specific expression, and the weight of the action expression in a document including the action expression;
Based on the similarity between the weight of one or more of the behavioral expressions that are connected to the unknown word and the weight of the one or more of the behavioral expressions that are connected to the specific expression, the unknown A semantic estimator that estimates the meaning of the word;
A search method storage unit that stores the behavioral expression and a search method in association with each other;
The specific expression storage unit identifies the action expression associated with the meaning estimated by the meaning estimation unit, and the search method storage unit includes the search method associated with the action expression. A search method selection section to select;
A search unit for searching for the unknown word by the search method selected in the search method selection unit;
An action recommendation device comprising: an output unit that outputs a search result obtained by the search unit.
The specific expression storage unit stores the class to which the specific expression belongs and the weight in association with each other, stores the calculation result of the weight of the specific expression belonging to the class as the weight for the class,
The semantic estimation unit is connected to the weight of one or more of the behavioral expressions that are connected to the unknown word and to one or more of the specific expressions stored in the specific expression storage unit. When the similarity with the weight of the action expression is smaller than a predetermined threshold, the weight of the one or more action expressions connected to the unknown word and the 1 or 2 belonging to the class The behavior recommendation device according to claim 1, wherein the meaning of the class to which the unknown word belongs is estimated as the meaning of the unknown word based on the similarity with the weight of the behavior expression.
The behavior according to claim 2, wherein the meaning estimation unit determines the class having the highest similarity as a class to which the unknown word belongs, and estimates the meaning of the class as the meaning of the unknown word. Recommendation device.
The class is structured hierarchically,
The specific expression storage unit stores the calculation result of the weight of the specific expression belonging to each of the plurality of classes included in the class of the higher hierarchy as the weight of the class of the higher hierarchy than the class,
The semantic estimation unit includes the weights of one or more of the behavioral expressions connected to the unknown word, and the one or more of the behavioral expressions of the class stored in the specific expression storage unit. When the similarity with a weight is smaller than a predetermined threshold, the weight of one or more of the action expressions connected to the unknown word and the one or more of the classes of the higher hierarchy The behavior recommendation device according to claim 2 or 3, wherein the meaning of the class in the higher hierarchy to which the unknown word belongs is estimated as the meaning of the unknown word based on a similarity to the weight of the behavior expression. .
An unknown word storage unit that stores the unknown word, the action expression of the unknown word, and the weight of the action expression in association with each other;
When the unknown word extracted by the unknown word extraction unit is stored in the unknown word storage unit, the weight specifying unit is configured to associate the behavior associated with the unknown word in the unknown word storage unit. The behavior recommendation device according to claim 1, wherein an expression and the weight are specified.
The unknown word extraction unit extracts the unknown word from the content that the user is browsing,
Further comprising an unknown word storage unit that stores user identification information for identifying a user, the unknown word, the action expression of the unknown word, and the weight of the action expression in association with each other,
The weight specifying unit extracts the unknown word from content browsed by a predetermined user by the unknown word extracting unit, and the extracted unknown word and the user identification information of the predetermined user are the unknown When stored in association with each other in a word storage unit, the unknown word storage unit identifies the user identification information of the predetermined user and the action expression and the weight associated with the unknown word The behavior recommendation device according to claim 1.
A situation information specifying unit for specifying situation information indicating a situation when the content is being browsed;
The specific expression storage unit further stores the situation information in association with the specific expression and the weight,
The semantic estimation unit estimates the meaning based on the weight of the behavior expression associated with the situation information identified by the situation information identification unit in the specific expression storage unit. The behavior recommendation device according to claim 1.