JP2011048558A - Device and method for generating task model - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a task model generation device and a task model generation method for automatically constructing a task model. <P>SOLUTION: A domain designation part 101 accepts the input of a domain from a user, and a domain task retrieval pattern generation part 102 generates a retrieval pattern for domain task obtaining including the domain, and a Web retrieval part 106 retrieves a Web page including the generated retrieval pattern, and a retrieval pattern matching part 107 extracts a series of character string ("viewing a movie") including a retrieval pattern character string (for example, "movie") shown by a retrieval pattern and an operation character string (for example, "viewing") connected to the retrieval pattern character string for showing the operation as a domain task. A storage part 110 stores the extracted domain task and the access destination information of the Web page in association as a task model. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a task model generation apparatus and a task model generation method for generating a task model for accessing a service such as content on a network.

  2. Description of the Related Art Conventionally, there are search methods such as directory search and keyword search as a method for a mobile terminal user to solve problems using various services provided on a network. In such a search method, the user needs to search for a service applicable to solving the problem. For this search, the user selects a candidate from a directory hierarchy classified by a general category or a list of information obtained by keyword search, and determines the content of the candidate. Had to repeat until a service to have was found.

  However, the mobile terminal has limitations such as low processing capability for display, communication, and the like compared to a normal PC, and that the input operation is not easy due to character input using the numeric keypad. Such a restriction is a heavy burden for the user to verify the functions provided by various services. For this reason, it is difficult for a user who does not know in advance what kind of service exists on the network to find a service applicable to problem solving. In order to solve such difficulties, in Patent Document 1, a problem that is expected to be required by the user and a sub-problem generated in the process of solving the problem are prepared in advance on the server side. Navigate to a service that can solve the problem.

JP 2003-333831 A

  By the way, in Patent Document 1, a user manually analyzes a problem structure and constructs a problem model (hereinafter referred to as a task model). However, in order to construct a task model in an unknown new place or theme (domain). Requires a long period of construction and scrutiny by experts. The task model is a database in which the problem structure of the user in the real world is analyzed and the problem solving action (task) of the user is structured. In the task model, for each place (domain) in the real world, tasks that occur in relation to the place are structured in a tree shape. A part of the real world behavior model in the movie theater domain is shown in FIG. The relationship between tasks is expressed as an achievement relationship is-achieved-by. The upper task expresses abstract requests and problems of the user, and the lower task expresses more specific solutions.

  For example, in FIG. 15, a department store is a domain, and an abstract request for the domain (department store) includes selecting a department store, buying a product, going to a department store, and the like as higher-level tasks. In addition, the subordinate tasks include checking sale information and checking the business date and time, and in the subordinate tasks, checking event information and checking closed days.

  Therefore, in order to solve the above-described problems, an object of the present invention is to provide a task model generation device and a task model generation method that can automatically construct a task model.

  In order to solve the above-described problems, a task model generation device according to the present invention includes an input unit that receives an input of a character string from a user and a domain task search that generates a search pattern including the input character string received by the input unit. A pattern generation unit, a domain task search unit that searches for a Web page including the search pattern generated by the domain task search pattern generation unit, and a Web page searched by the domain task search unit. Search pattern character string, a domain task extraction unit that extracts a series of character strings including an operation character string that represents an operation in association with the search pattern character string, and a domain extracted by the domain task extraction unit Task and domain task included And a storage means for storing as a task model in association with the access destination information eb page.

  Further, the task model generation method of the present invention includes an input step for receiving an input of a character string from a user, a domain task search pattern generation step for generating a search pattern including the input character string received by the input step, A search step for searching for a Web page including the search pattern generated by the domain task search pattern generation step, a search pattern character string indicated by the search pattern from the Web page searched by the search step, and the search pattern A domain task extraction step for extracting a series of character strings including an operation character string connected to a character string and representing an operation as a domain task, a domain task extracted by the domain task extraction step, and a web page including the domain task No And a storage step of storing in the storage means as a task model in association with Seth destination information.

  With this configuration, a search pattern including an input character string received from a user is generated, a Web page including the generated search pattern is searched, and a search pattern character indicated by the search pattern is searched from the searched Web page. A sequence and a series of character strings including an operation character string representing an operation in association with the search pattern character string are extracted as a domain task. The extracted domain task and the access destination information of the Web page including the domain task are associated with each other and stored as a task model. Thereby, the task model (database) for user's action support can be generated. For example, it can be expected that a task model is constructed in a few hours in the present invention, whereas it takes about one month due to manual work or the like.

  Further, in the task model generation device of the present invention, the domain task search pattern generation unit includes a particle for linking to an input character string received by the input unit to be connected to a word representing an action, and the input character string It is preferable to generate a search pattern by concatenating the back of the search pattern.

  With this configuration, it is possible to generate a search pattern by concatenating a particle, which is linked to a word representing an action by connecting to an accepted input character string, behind the input character string. Thereby, the character string for supporting a user's action can be represented exactly, and the task model for it can be produced | generated.

  In the task model generation device of the present invention, it is preferable that the domain task search pattern generation unit generates a search pattern by connecting the particle “” to the end of the input character string.

  With this configuration, it is possible to generate a search pattern by connecting the particle "" to the back of the input character string, and to accurately represent the character string for supporting the user's action, and a task model for that purpose. Can be generated.

  In the task model generation device of the present invention, the domain task extraction unit performs dependency analysis to determine an operation character string corresponding to the input character string, and includes the search pattern and the operation character string. It is preferable to extract a series of character strings.

  With this configuration, by performing dependency analysis, it is possible to determine an operation character string corresponding to an input character string and extract a series of character strings including a search pattern and an operation character string. Thereby, the character string for supporting a user's action can be represented exactly, and the task model for it can be produced | generated. In other words, a more accurate task model can be constructed by extracting a character string considering the meaning of a sentence, regardless of character pattern matching.

  In the task model generation device of the present invention, it is preferable that the domain task extraction unit extracts a character string composed of only kanji characters representing an operation as a domain task together with the search pattern. .

  With this configuration, a character string that is connected immediately after the search pattern and consists of only kanji characters representing an action is extracted as a domain task together with the search pattern, thereby supporting a user's action without using a particle. Can be accurately represented, and a task model for that can be generated.

  In addition, the task model generation device of the present invention connects a domain task character string, which is a domain task extracted by the domain task extraction unit, to a modifier for modifying an operation indicated by the domain task character string. An included task search pattern generating means for generating an included task search pattern to which a concatenated character string to be added is added, wherein the domain task search means uses the included task search pattern generated by the included task search pattern generating means. It is preferable that the contained Web page is searched, and the domain task extraction unit extracts an included task character string in which a modifier character string indicating a modifier is connected to the included task search pattern.

  With this configuration, an inclusive task search pattern is generated by adding a concatenated character string for concatenating a domain task character string that is a domain task to a modifier for modifying the operation indicated by the domain task character string. Then, the generated Web page including the included task search pattern is searched, and the included task character string in which the modifier character string indicating the modifier is connected to the included task search pattern can be extracted. As a result, tasks including domain tasks can be acquired, and actions for user support can be further subdivided and supported.

  The task model generation device of the present invention further includes a domain co-occurrence degree calculating unit that calculates a co-occurrence degree indicating a degree of association between the input character string and the action character string, and the domain task extracting unit includes the domain task extracting unit, It is preferable to extract a domain task having a co-occurrence degree equal to or greater than a predetermined value based on the calculation result by the domain co-occurrence degree calculating means.

  With this configuration, it is possible to calculate the co-occurrence degree indicating the degree of association between the input character string and the action character string, and to extract a domain task having a co-occurrence degree equal to or greater than a predetermined value based on the calculation result. As a result, character strings having a high degree of association can be linked, more accurate action support can be provided to the user, and unrelated web pages can be removed.

  The task model generation device of the present invention further includes an inclusion task co-occurrence degree calculating means for calculating a co-occurrence degree indicating a degree of association between the domain task and the modifier character string, and the domain task extracting means includes the It is preferable to extract an inclusion task having a co-occurrence degree equal to or greater than a predetermined value based on the calculation result by the inclusion task co-occurrence degree calculation means.

  With this configuration, it is possible to calculate the co-occurrence degree indicating the degree of association between the domain task and the modification character string, and to extract an inclusion task having a co-occurrence degree equal to or greater than a predetermined value based on the calculation result. As a result, character strings having a high degree of association can be linked, more accurate action support can be provided to the user, and unrelated web pages can be removed.

  In addition, the task model generation device of the present invention includes an input unit that receives an input of a character string from a user, a subdomain search pattern generation unit that generates a search pattern including an input character string received by the input unit, From a sub-domain search means for searching a Web page including a search pattern generated by the sub-domain search pattern generation means, and from a noun character string linked to the search pattern from the Web page searched by the domain task search means Subdomain extraction means for extracting a series of character strings as subdomains, and related task search pattern generation means for generating as a related task search pattern from a subdomain character string indicating a subdomain extracted by the subdomain extraction means; , Related task search pattern generation Related task search means for searching for a Web page including the search pattern generated by the stage, and operations associated with the search pattern character string and the search pattern character string from the Web page searched by the related task search means The related task extracting means for extracting a series of character strings consisting of action character strings representing the related task, the related task extracted by the related task extracting means, and the access destination information of the Web page including the related task are associated with each other. Storage means for storing the information.

  In addition, the task model generation method of the present invention includes an input step of receiving a character string input from a user, a subdomain search pattern generation step of generating a search pattern including the input character string received by the input step, From a sub-domain search step for searching a Web page including a search pattern generated by the sub-domain search pattern generation step, and from a noun character string linked to the search pattern from the Web page searched by the sub-domain search step A subdomain extraction step for extracting a series of character strings as subdomains, and a related task search pattern generation step for generating a subdomain character string indicating the subdomain extracted by the subdomain extraction step as a search pattern for the related tasks; A related task search step for searching a Web page including the search pattern generated by the related task search pattern generation step, and the search pattern character string and the search pattern character from the Web page searched by the related task search step A related task extraction step for extracting a series of character strings composed of operation character strings representing operations connected to the column as related tasks, a related task extracted by the related task extraction step, and a Web page including the related tasks A storage step of associating the access destination information with the storage means.

  With this configuration, a search pattern including an input character string that is accepted by the user as a character string is generated, and a Web page including the generated search pattern is searched. Then, a series of character strings composed of noun character strings linked to the search pattern is extracted as subdomains from the searched Web page, and a subdomain character string indicating the extracted subdomain is generated as a search pattern. A search is made for a Web page including the generated search pattern, and a series of character strings consisting of a search pattern character string and an operation character string representing an operation linked to the search pattern character string are searched from the searched Web page. Extract as The extracted related task and the access destination information of the Web page including the related task are stored in association with each other. Thereby, the task model (database) for user's action support can be generated. For example, it can be expected that a task model is constructed in a few hours in the present invention, whereas it takes about one month due to manual work or the like.

  The task model generation device of the present invention includes a domain task search pattern generation unit that generates a search pattern including an input character string received by the input unit, and a search pattern generated by the domain task search pattern generation unit. A domain task search means for searching for a Web page including the search pattern character string, a search pattern character string indicated by the search pattern from the Web page searched by the domain task search means, and an operation associated with the search pattern character string A domain task extraction unit that extracts a series of character strings including an operation character string as a domain task, and the storage unit includes a domain task extracted by the domain task extraction unit and a Web page including the domain task. Matching access destination information A task model generation device for storing as a model, the related task co-occurrence degree calculating means for calculating a co-occurrence degree indicating a degree of association between a domain task consisting of a predetermined character string and the related task; It is preferable to include association processing means for associating the related task with the domain task based on the co-occurrence degree calculated by the task co-occurrence degree calculating means.

  With this configuration, the co-occurrence degree indicating the degree of association between the domain task consisting of a predetermined character string and the related task is calculated, and the association between the related task and the domain task is performed based on the calculated co-occurrence degree. Do. As a result, the related task can be associated with a higher related domain task, and the user's action support can be performed more accurately.

  ADVANTAGE OF THE INVENTION According to this invention, the task model (database) for a user's action assistance can be produced | generated. For example, it can be expected that a task model is constructed in a few hours in the present invention, whereas it takes about one month due to manual work or the like.

1 is a system configuration diagram illustrating a system configuration of a communication system 10 including a task model generation device 100 of the present embodiment. 3 is a block diagram illustrating functions of a task model generation device 100. FIG. 2 is a hardware configuration diagram of a task model generation device 100. FIG. It is explanatory drawing which shows the process in which a domain task is extracted. It is explanatory drawing which shows the processing process until an inclusion task is extracted. The schematic diagram which shows the relationship of a domain, a domain task, and an inclusion task is shown. 4 is a flowchart illustrating a method for extracting a domain task of the task model generation device 100. It is a flowchart which shows the process of the task model production | generation apparatus 100 which produces | generates an inclusion task. It is a block diagram which shows the function of the task model production | generation apparatus 100a for extracting a related task. It is explanatory drawing which shows the process of extracting a related task. It is a flowchart which shows the process of the task model production | generation apparatus 100a of this embodiment. It is a schematic diagram which shows the correlation of a domain task and a related task. It is explanatory drawing explaining the memory | storage form of a task model. It is explanatory drawing which shows the screen which displays a task model. It is a conceptual diagram which shows the concept of a task model.

  Embodiments of the present invention will be described with reference to the accompanying drawings. Where possible, the same parts are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a system configuration diagram showing a system configuration of a communication system 10 including a task model generation device 100 of the present embodiment. The communication system 10 includes a task model generation device 100 and a search engine 200. In this communication system 10, the task model generation device 100 uses a search engine 200 to search for a Web page that includes a predetermined character string, and corresponds the searched Web page to a domain task that is a predetermined character string. It is memorized as the attached task model.

  Next, the task model generation device 100 will be described. FIG. 2 is a block diagram illustrating functions of the task model generation device 100. As shown in FIG. 2, the task model generation device 100 includes a domain designation unit 101 (input unit), a domain task search pattern generation unit 102 (domain task search pattern generation unit), and an included task search pattern generation unit 103 (included task search). Pattern generation means), Web search section 106 (domain task search means), search pattern matching section 107 (domain task extraction means), co-occurrence degree calculation section 108 (domain co-occurrence degree calculation means, inclusion task co-occurrence degree calculation means) The task model generation unit 109 and the storage unit 110 (storage means) are included.

  As shown in FIG. 3, the task model generation apparatus 100 is realized by hardware such as a CPU and a RAM. FIG. 3 is a hardware configuration diagram of the task model generation device 100. As shown in FIG. 3, the task model generation device 100 shown in FIG. 2 physically includes a CPU 11, a RAM 12 and a ROM 13 that are main storage devices, an input device 14 such as a keyboard and a mouse that are input devices, a display, and the like. Output device 15, a communication module 16 which is a data transmission / reception device such as a network card, an auxiliary storage device 17 such as a hard disk, and the like. Each function described in FIG. 2 has the input device 14, the output device 15, and the communication module 16 under the control of the CPU 11 by reading predetermined computer software on the hardware such as the CPU 11 and the RAM 12 shown in FIG. 3. This is realized by reading and writing data in the RAM 12 and the auxiliary storage device 17. Hereinafter, each functional block will be described based on the functional blocks shown in FIG.

  The domain designating unit 101 is a part that receives an input of a character string (input character string) that is a domain from an operator of the task model generation device 100. For example, in the example of FIG. 14, “department store” corresponds to a domain. This domain is a character string for defining the area of the task model to be learned, and is a part indicating the core of the task model for supporting the user's action.

  The domain task search pattern generation unit 102 is a part that generates a search pattern based on an input character string indicating a domain input by the domain specification unit 101. For example, the domain task search pattern generation unit 102 generates a search pattern for acquiring a domain task that is a task directly under the domain, and generates a search pattern composed of “domain + particle”. As typical particles, there are “no”, “de”, “ha”, “ha”, and the like. Here, “ha” is selected as a particle constituting the search pattern. The domain task search pattern generation unit 102 uses, for example, “movie” as a search pattern, and the Web search unit 106 to be described later can search for a Web page including “movie”. Thereby, for example, domain tasks such as “watch movie” and “make movie” can be acquired. The domain task is a character string indicating the user's action and includes a verb indicating the user's action or a noun indicating the action.

  The Web search unit 106 is a part that accesses the search engine 200 and performs a search using the search pattern generated by the domain task search pattern generation unit 102. For example, the Web search unit 106 is a part that causes the search engine 200 to search for a search pattern composed of a character string “movie” generated by the domain task search pattern generation unit 102.

  The search pattern matching unit 107 is a part that extracts a domain task (character string) that matches a predetermined lexical syntactic pattern from the snippet (search result) of the Web page searched by the Web search unit 106. For example, when “domain“ movie ”+ particle“ to ”+ verb” is set as the lexical syntactic pattern, the search pattern matching unit 107 sets sentences (task (verb The character string shown)) is extracted using dependency analysis or the like. As a result, the search pattern matching unit 107 can extract a Web page including a domain task (character string) such as “watch movie” or “create movie” as a task model. Here, by performing dependency analysis, which is a well-known technique, without extracting only a series of character strings such as “watch movie”, a character string such as “watch movie at movie theater” is also included. Can be extracted.

Note that lexico-syntactic patterns (lexico-syntactic patterns) is a typical technique for extracting knowledge from a large-scale text corpus. From a specific expression, lexico-syntactic patterns (Hyponymy Relation) and conceptual concrete relationship ( It is widely used as a method for obtaining concept-instance relations (see references below).
[Reference 1]
[1] Philipp Cimiano: Ontology Learning And Population from Text: Algorithms, Evaluation And Applications, Springer-Verlag, 2006.
[Reference 2]
[2] David Sanchez: Domain Ontology Learning from the Web, VDM Verlag, 2008.

  Regarding the domain task search pattern generation unit 102 and the search pattern matching unit 107, the following modifications may be possible. That is, in the above example, it is assumed that a particle is used, but a particle is not necessarily required. For example, if the domain is “movie”, the search pattern is “movie”, and the lexical syntactic pattern is “domain + a noun consisting only of kanji representing actions (for example, appreciation or production)”. The pattern generation unit 102 uses the domain as a search pattern as it is, and the search pattern matching unit 107 generates a series of character strings in which nouns representing actions are added to the domain (for example, “movie”) according to the lexical syntax pattern. You may make it extract.

ここで、Hits(X)とは、検索エンジンに文字列Ｘを入力したときの検索数である。
［参考文献３］
[3]P.D.Turney: Mining the Web for Synonyms: PMI-IR versus LSA on TOEFL, ECML2001,pages 491-502, 2001. The co-occurrence degree calculation unit 108 indicates the degree of association between the character string indicating the domain specified by the domain specifying unit 101 and the task (the verb part character string) in the lexical syntactic pattern extracted by the search pattern matching unit 107. This is the part that calculates the co-occurrence. Since only the processing by the search pattern matching unit 107 includes noise, it is intended to filter only appropriate concepts by calculating the co-occurrence degree. Here, the calculation of the degree of co-occurrence is a method of determining the upper and lower relations between concepts proposed by Reference Document 3 below from the number of web page searches. The evaluation value (Score) indicating the degree of inclusion of the subordinate concept B with the superordinate concept A is expressed by the following equation.

Here, Hits (X) is the number of searches when the character string X is input to the search engine.
[Reference 3]
[3] PDTurney: Mining the Web for Synonyms: PMI-IR versus LSA on TOEFL, ECML2001, pages 491-502, 2001.

  For example, when the co-occurrence degree calculation unit 108 wants to calculate the co-occurrence degree of “movie” and “create”, the number of searches for the Web page including “movie” and the number of Web pages including “create” The co-occurrence degree is calculated based on the number of searches. When the task model generation unit 109 determines that the threshold value T is equal to or greater than the threshold value T, the two character strings are treated as having a high degree of association and used as a domain task.

  The task model generation unit 109 stores the domain task extracted by the search pattern matching unit 107 as a task model in which the co-occurrence degree calculated by the co-occurrence degree calculation unit 108 is equal to or greater than the threshold T in the storage unit 110. It is a part to be memorized. For example, this threshold value T is preferably set to 0.00001 from an empirical rule.

  Further, the task model generation unit 109 outputs a domain task character string indicating the generated domain task to the inclusion task search pattern generation unit 103. The processing in the included task search pattern generation unit 103 will be described later.

  The storage unit 110 is a part that stores the task model generated by the task model generation unit 109. For example, the storage unit 110 is a management table shown in FIG. 13 and stores a task ID, a task name, and a parent task ID in association with each other. In FIG. 13, task ID: 1001 describes a task name: “watch movie” and its parent task ID: 0, which indicates a domain task. In addition, for a part described as task ID: 1002, task name: “watch movie in cinema”, parent task ID: 1001, parent task ID: 1001, lower layer of task name “watch movie” This is a part indicating an inclusion task. In FIG. 13, although not shown, a URL for accessing the Web page is associated with each task.

  These task IDs are assigned uniquely by the task model generation unit 109, and the parent task ID is a corresponding domain task when the domain task is associated with the inclusion task. Are assigned to the parent task ID column of the included task.

  Using the information stored in the storage unit 110, the user (mobile terminal 300) can perform screen display as shown in FIG. That is, the user (mobile terminal 300) accessing the information stored in the storage unit 110 operates the pointer 303 and presses the service acquisition button 301 while looking at the screen 302, thereby the domain task. A web page associated with an inclusion task or related task can be accessed. Therefore, when the user wants to watch a movie in a movie theater, if the item “Watch a movie in a movie theater” is selected, a list of Web pages related to watching a movie is displayed. By selecting one, it is possible to browse a web page.

  FIG. 4 shows a process of extracting a domain task. FIG. 4 shows an example of a search pattern for domain task acquisition, an input example input to the search engine, a search result, and a co-occurrence degree calculation result. As shown in FIG. 4, here, “movie” is generated as a search pattern, and this character string is input to the search engine 200, and search processing is performed. As a search result, a snippet in which the character string “movie” is highlighted in bold is obtained as a snippet. In FIG. 4, only one search result is shown, but tens of thousands of search results are usually obtained.

  Then, from this search result, the character string of the verb that leads to “movie” is extracted. Here, “create” is extracted as a verb string. By performing dependency analysis, several verb character strings are extracted in addition to “create”. Then, the co-occurrence degree calculation is performed, and here, the co-occurrence degree of “movie” and “make” is calculated. Here, an AND search for “movie” and “create” is performed, and a search for only “create” is performed, and the number of searches for 12.3 million and 1.43 million is retrieved, respectively. Is calculated as 0.86. Since the co-occurrence degree 0.86 is larger than the threshold value T, “create movie” is extracted as a domain task, and the Web page (or URL) related thereto is associated with the domain task as a task model and is stored in the storage unit 110. Is remembered.

  The included task search pattern generation unit 103 is a part that generates a search pattern for extracting an included task including the domain task acquired as described above. An inclusion task is a task including a domain task, and is a character string composed of “modifier + particle + domain task”. In other words, the inclusion task indicates an action belonging to the lower level of the domain task. Here, the inclusion task search pattern generation unit 103 selects “de” as a particle and generates a search pattern such as “watch a movie”, and the Web search unit 106 can search for this search pattern. . In the present embodiment, a character string that is linked to a modifier for modifying the operation of a domain task such as the particle “de” is referred to as a connection character string.

  On the other hand, the lexical syntactic pattern becomes “modifier + inclusion task” obtained by adding a modifier to the inclusion task, and the search pattern matching unit 107 extracts a character string having such a lexical syntactic pattern. Note that by performing dependency analysis, the search pattern matching unit 107 can also extract the case where the particle “by” is separated, such as “... by watching a movie”.

  The Web search unit 106 searches for a Web page including the included task search pattern generated by the included task search pattern generation unit 103, and the search pattern matching unit 107 searches for a character string (modifier) having a predetermined lexical syntactic pattern. Word + particle + domain task).

  Then, the co-occurrence degree calculation unit 108 calculates the co-occurrence degree between the domain task and the modifier, and the task model generation unit 109 stores the inclusion task having the co-occurrence degree equal to or greater than the threshold T in the storage unit 110 as a task model. Remember.

  This inclusion task extraction will be described with reference to FIG. FIG. 5 is an explanatory diagram showing a processing process until an inclusion task is extracted. In FIG. 5, the inclusive task acquisition search pattern is set to “watch movie with”. And this is input into a search engine, and a search result is obtained. As in FIG. 4, the character string corresponding to this inclusion task acquisition search pattern is shown in bold. Similar to FIG. 4, tens of thousands of Web pages are searched as search results.

  Then, the co-occurrence degree is calculated. In FIG. 5, the degree of co-occurrence between “watch movie” and “DVD” and the degree of co-occurrence between “watch movie” and “movie theater” are calculated.

  Next, a process for extracting a domain task of the task model generation device 100 configured as described above will be described. FIG. 7 is a flowchart illustrating a method for extracting a domain task of the task model generation device 100. As shown in FIG. 7, a domain (character string) is input and specified from the domain specifying unit 101 by the operator (S101). Then, a search pattern for extracting a domain task is generated by the domain task search pattern generation unit 102 (S102).

  Next, the Web search unit 106 performs a search using the search pattern generated by the domain task search pattern generation unit 102 (S103). Here, when the search by all the search patterns is executed (S104: YES), the character string matching the search pattern is extracted by the search pattern matching unit 107 (S105). Then, the co-occurrence degree calculation unit 108 calculates the co-occurrence degree indicating the degree of association between the domain and the character string (S106). Then, the task model generation unit 109 selects a character string having a co-occurrence degree equal to or greater than the threshold T, and generates a domain task including the character string (S107). The generated domain task is stored in the storage unit 110 together with the URL, and is held accessible by the user (S108).

  In this manner, a corresponding character string is extracted according to a search pattern for domain search based on the designated domain, and a domain task including the character string is generated.

  Next, processing for generating an inclusion task will be described. FIG. 8 is a flowchart showing the processing of the task model generation device 100 that generates an inclusion task. First, a domain task is generated by the method shown in FIG. 7 (S201). Then, the search pattern for extracting the inclusion task is generated by the inclusion task search pattern generation unit 103 (S202). Then, the Web search unit 106 performs a search using the search pattern generated by the inclusion task search pattern generation unit 103 (S203). If a search using all search patterns is executed (S204: YES), a character string that matches the search pattern is extracted by the search pattern matching unit 107, and a character string that matches the lexical syntactic pattern is extracted. (S205). Then, the co-occurrence degree calculation unit 108 calculates the co-occurrence degree indicating the degree of inclusion between the domain task and the modifier character string indicating the modifier of the character string (S206). Then, the task model generation unit 109 selects a modified character string having a co-occurrence degree equal to or greater than the threshold T, and an included task including the modified character string is generated (S207). The generated inclusion task is stored in the storage unit 110 together with the URL, and is held accessible by the user (S208).

  In this way, a domain is specified, a domain task based on the specified domain is extracted, and an inclusion task that includes this domain task is extracted. FIG. 6 is a schematic diagram showing the relationship between domains, domain tasks, and inclusion tasks. As shown in FIG. 6, “movie” is designated as a domain, and “watch movie”, “appear in movie”, and “create movie” are extracted as domain tasks based on this domain. In addition, “watch movie at cinema”, “watch movie at home”, “watch movie with subtitles”, “watch movie on TV” are extracted as inclusion tasks included in “watch movie” The In this way, a task model for associating a domain, a domain task, and an inclusion task and making it possible to support user behavior can be generated.

  Next, related tasks related to the domain task will be described. This related task is a task closely related to the domain task among the character strings including the domain. For example, when the domain is “movie” and the domain task is “watch movie”, the related task is “buy a movie ticket” which is closely related to the domain task “watch movie”. Note that deeply related means that the co-occurrence degree between a domain task and a related task is a predetermined value or more. The detailed processing will be described below.

  FIG. 9 is a block diagram illustrating functions of the task model generation device 100a for extracting related tasks. The task model generation device 100a includes a domain designation unit 101 (input unit), a domain task search pattern generation unit 102 (domain task search pattern generation unit), an included task search pattern generation unit 103, a Web search unit 106 (domain task search unit). ), Search pattern matching unit 107 (domain task extraction unit), co-occurrence degree calculation unit 108, task model generation unit 109, storage unit 110, subdomain search pattern generation unit 111 (subdomain search pattern generation unit), Web search unit 112 (subdomain search means), search pattern matching section 113 (subdomain extraction means), related task search pattern generation section 115 (related task search pattern generation means), co-occurrence degree calculation section 117 (related task co-occurrence degree calculation means) ) And the association processing unit 118 ( Configured to include a communication with the processing unit).

  The task model generation device 100a is configured by hardware including a CPU and the like. For example, various functions are realized by the hardware illustrated in FIG. Hereinafter, each functional block will be described. The domain specifying unit 101, the domain task search pattern generation unit 102, the inclusion task search pattern generation unit 103, the Web search unit 106, the search pattern matching unit 107, the co-occurrence degree calculation unit 108, the task model generation unit 109, and the storage unit 110 Has the same functional configuration as that of the task model generation device 100, and a detailed description thereof will be omitted.

  The subdomain search pattern generation unit 111 is a part that generates a search pattern for a subdomain. The sub-domain is a character string including the domain input from the domain designating unit 101. In the present embodiment, the sub-domain is a character string set as “domain + particle”. For example, if the domain is “movie”, the subdomain is “movie”.

  The Web search unit 112 is a part that accesses the search engine 200 and performs a search process based on the subdomain search pattern generated by the subdomain search pattern generation unit 111. For example, the Web search unit 112 is a part that causes the search engine 200 to search for a search pattern composed of the character string “movie” generated by the subdomain search pattern generation unit 111.

  The search pattern matching unit 113 is a part that extracts a subdomain (character string) that matches a predetermined lexical syntactic pattern from a snippet (search result) of a Web page searched by the Web search unit 112. For example, when “domain“ movie ”+ particle“ no ”+ noun” is set as the lexical syntactic pattern, the search pattern matching unit 113 performs dependency analysis on the sentences constituting ““ movie ”+ noun”. Extract using etc. As a result, the search pattern matching unit 113 can extract a Web page including a subdomain (character string) such as “movie ticket” and “movie schedule” as a subdomain. Here, by performing dependency analysis, which is a well-known technique, without extracting only a series of character strings such as “movie ticket”, for example, a character string such as “movie preview ticket” is also included. Can be extracted. Further, the search pattern matching unit 113 may extract a character string according to a lexical syntactic pattern “domain + noun” as a subdomain. That is, it may be a character string that does not use a particle, and in the above example, “movie ticket” may be a subdomain.

  Further, the search pattern matching unit 113 matches a predetermined lexical syntactic pattern from a search result (snippet) searched based on a search pattern of related tasks generated by the related task search pattern generation unit 115 described later. Related tasks (character strings) are extracted. Here, a related task (character string) including a subdomain is extracted, and a related task having “subdomain + particle + verb” as a lexical syntactic pattern is extracted. For example, a related task composed of the subdomain “movie ticket”, the particle “has”, and the verb “buy” is extracted. The particle “in” may be replaced with “no”, “de”, etc., and other particles may be applied.

  The related task search pattern generation unit 115 is a part that generates a search pattern for related tasks based on the subdomain extracted by the search pattern matching unit 113. The related task is a character string including a subdomain. In this embodiment, the related task is a character string set as “subdomain + particle”. For example, when the subdomain is “movie ticket”, the search pattern is “movie ticket”. The Web search unit 112 searches for a Web page including the search pattern, and the search pattern matching unit 113 extracts a related task having “domain + particle” as “+ verb” as a lexical syntactic pattern. For example, the search pattern matching unit 113 can extract a related task such as “purchase a movie ticket”. In some cases, particles are not used, and in this case, a related task for “movie ticket purchase” may be extracted.

  The co-occurrence degree calculation unit 117 is a part that calculates the co-occurrence degree indicating the degree of association between the domain task generated in the task model generation unit 109 and the extracted related task in the search pattern matching unit 113. That is, the co-occurrence degree calculation unit 117 determines a domain task to which one related task is associated with a plurality of domain tasks generated in the task model generation unit 109. Calculate the co-occurrence between each of the. A specific method for calculating the co-occurrence degree is the same as that shown in Equation 1.

  The association processing unit 118 performs processing for associating one related task with the domain task having the largest co-occurrence value based on the co-occurrence degree calculated by the co-occurrence degree calculation unit 117. In other words, the association processing unit 118 determines which domain task is to store the related task in the storage unit 110 and stores the related task in association with the domain task. Specifically, as shown in FIG. 13, the association processing unit 118 performs association processing by describing the task ID of the associated domain task in the parent task ID column of the related task. The association processing unit 118 assigns the task ID of the related task to be unique.

  A process until the related task is extracted in the task model generation device 100a configured as described above will be described with reference to FIG. FIG. 10 is an explanatory diagram showing a process of extracting related tasks.

  FIG. 10 shows that the search pattern “movie” is generated as the subdomain acquisition search pattern. Then, using this, the search engine 200 is caused to search for “movie”. In the search engine 200, for example, a search result as shown in the search result column of FIG. 10 is obtained. For example, “movie ticket” or “movie English” can be obtained as a subdomain.

  Next, an achievement task acquisition search pattern is generated, and a search based on this is performed. For example, in this embodiment, 1) “movie ticket” and 2) “movie movie” are generated, and a search based on this is performed.

  In the calculation of the co-occurrence, if “watch movie” and “learn with movie” are generated as domain tasks, the co-occurrence between each domain task and the related task “buy movie ticket” The degree of occurrence is calculated by the co-occurrence degree calculation unit 117. In this embodiment, the co-occurrence of the related task “purchase a movie ticket” and the domain task “watch a movie” is 0.052, the related task “purchase a movie ticket”, and the domain task “ Since the co-occurrence with “learn in movies” is 0.0021, the related task “buy movie tickets” is associated with the domain task “watch movies”. Similarly, the related task “listening to English of the movie” is also associated with the front task.

  Next, processing of the task model generation device 100a configured as described above will be described. FIG. 11 is a flowchart showing the processing of the task model generation device 100a of this embodiment. First, a task model for the domain task is generated by the task model generation unit 109 (S301). This detailed processing is as shown in FIG. 7 and FIG.

  Next, a subdomain search pattern is generated by the subdomain search pattern generation unit 111 (S302), and a Web page including the search pattern of the subdomain is searched by the Web search unit 112 (S303). When the Web search unit 112 determines that all search patterns have been executed (S304), the search pattern matching unit 113 extracts subdomains that match the search pattern (S305).

  The related task search pattern generation unit 115 generates a search pattern for related tasks using the subdomain extracted by the search pattern matching unit 113 (S306). Then, the search by the search pattern of the related task is performed by the Web search unit 112 (S307). If the Web search unit 112 determines that all search patterns have been executed (S308), the search pattern matching unit 113 extracts related tasks that match the search pattern (S309).

  When related tasks are extracted, to determine which domain task should be associated, the co-occurrence degree between one extracted related task and multiple domain tasks that have already been extracted is calculated. (S310). Then, the related task is associated with the domain task having a high co-occurrence degree (S311), and the association is stored in the storage unit 110 (S312).

  FIG. 12 shows a schematic diagram showing the association between domain tasks and related tasks processed in this way. As shown in FIG. 12, the domain task “watch movie” has associated tasks “watch movie at cinema”, “watch movie at home”, and the like associated with lower layers. The domain task “watch movie” is associated with related tasks “buy movie ticket”, “check movie schedule”, and the like.

  Next, operational effects of the task model generation device 100 and the task model generation device 100a of the present embodiment will be described. In the task model generation device 100 of the present embodiment, the domain specification unit 101 receives an input of a character string as a domain from a user, and the domain task search pattern generation unit 102 acquires a domain task including this domain (input character string). Generate a search pattern for The Web search unit 106 searches for a Web page including the generated search pattern, and the search pattern matching unit 107 searches the search pattern character string (for example, “movie”) indicated by the search pattern from the searched Web page. ), And a series of character strings (“watch movie”) including an action character string (for example, “watch”) representing an action linked to the search pattern character string are extracted as domain tasks. The storage unit 110 stores the extracted domain task and the access destination information (URL) of the Web page including the domain task in association with each other as a task model. Thereby, the task model (database) for user's action support can be generated. For example, it can be expected that a task model is constructed in a few hours in the present invention, whereas it takes about one month due to manual work.

  Further, in the task model generation device 100 of the present embodiment, a particle (this book) for linking to an input character string (for example, “movie”) that is a domain accepted by the domain designating unit 101 and linked to a phrase representing an action. In an embodiment, “but” (but not limited to) can be concatenated after the input string to generate a search pattern (eg, “movie”). Thereby, the character string for supporting a user's action can be represented exactly, and the task model for it can be produced | generated.

  Further, in the task model generation device 100 of the present embodiment, a particle can be generated by concatenating the particle "" to the back of the input character string as a domain, and generate a search pattern. Can be accurately represented, and a task model for that can be generated.

  In the task model generation device 100 of this embodiment, the search unit 106 or the search engine 200 performs dependency analysis, so that the search pattern matching unit 107 has an input character string (for example, “movie”) serving as a domain. Can be extracted, and a series of character strings including the search pattern and the action character string can be extracted. Thereby, the character string for supporting a user's action can be represented exactly, and the task model for it can be produced | generated. In other words, a more accurate task model can be constructed by extracting a character string considering the meaning of a sentence, regardless of character pattern matching.

  Further, in the task model generation device 100 of the present embodiment, the search pattern matching unit 107, when an input character string as a domain is used as it is as a search pattern, is connected immediately after the search pattern and only the kanji representing the operation. A character string for supporting a user's action without using a particle by extracting a character string consisting of (for example, “watching”) as a domain task (in this case, “watching movie”) together with a search pattern. The column can be accurately represented, and a task model for it can be generated.

  Further, in the task model generation device 100 of the present embodiment, the inclusion task search pattern generation unit 103 indicates a domain task character string (for example, “watch movie”) that is a domain task by the domain task character string. An inclusion task search pattern (for example, “watch a movie with”) to which a concatenated character string (for example, “by”) is added to a modifier for modifying the generated operation is generated. The Web search unit 106 searches for a Web page that includes the generated inclusion task search pattern, and the search pattern matching unit 107 uses a modifier character string (for example, “in a movie theater”) that indicates a modifier in the inclusion task search pattern. Can be extracted (for example, “watch a movie in a movie theater”). As a result, tasks including domain tasks can be acquired, and actions for user support can be further subdivided and supported.

  In the task model generation device 100 of the present embodiment, the co-occurrence degree calculation unit 108 calculates a co-occurrence degree indicating the degree of association between the input character string serving as the domain and the action character string. Based on the calculation result, the task model generation unit 109 can extract a domain task having a co-occurrence degree equal to or greater than a predetermined value. As a result, character strings having a high degree of association can be linked, more accurate action support can be provided to the user, and unrelated web pages can be removed.

  Further, in the task model generation device 100 of the present embodiment, the co-occurrence degree indicating the degree of association between the domain task generated by the task model generation unit 109 and the modified character string is calculated, and the task model generation unit 109 performs this calculation. Based on the result, an inclusion task having a co-occurrence degree equal to or greater than a predetermined value can be extracted. As a result, character strings having a high degree of association can be linked, more accurate action support can be provided to the user, and a web page with little relation can be provided.

  In the task model generation device 100a of this embodiment, the subdomain search pattern generation unit 111 generates a search pattern (for example, “movie”) including an input character string received from the user in the domain specification unit 101. Then, the Web search unit 112 searches for a Web page including the generated search pattern. Then, the search pattern matching unit 113 sub-domains a series of character strings (for example, “movie ticket”) composed of noun character strings (for example, “ticket”) linked to the search pattern from the searched Web page. The related task search pattern generation unit 115 generates a subdomain character string indicating the extracted subdomain as a search pattern for acquiring a related task (for example, “movie ticket”). The Web search unit 112 further searches for a Web page including the generated search pattern, and the search pattern matching unit 113 is linked to the search pattern character string and the search pattern character string from the searched Web page. A series of character strings consisting of action character strings representing actions (for example, “purchase”) are extracted as related tasks (for example, “purchase a movie ticket”). The storage unit 110 stores the extracted related task and the access destination information of the Web page including the related task in association with each other. Thereby, the task model (database) for user's action support can be generated. For example, it can be expected that a task model is constructed in a few hours in the present invention, whereas it takes about one month due to manual work.

  Further, in the task model generation apparatus 100a of the present embodiment, the co-occurrence degree calculation unit 117 includes a domain task (for example, “watch movie”) composed of a predetermined character string and a related task (for example, “movie ticket”). The co-occurrence degree indicating the degree of relevance to “purchased”) is calculated, and the related task and the domain task are associated based on the calculated co-occurrence degree. As a result, the related task can be associated with a domain task having a higher degree of association, and the user's action support can be performed more accurately.

DESCRIPTION OF SYMBOLS 10 ... Communication system, 100 ... Task model generation apparatus, 100a ... Task model generation apparatus, 101 ... Domain specification part, 102 ... Domain task search pattern generation part, 103 ... Inclusion task search pattern generation part, 104 ... Sub domain search pattern generation , 105 ... related task search pattern generation unit, 106 ... Web search unit, 107 ... search pattern matching unit, 108 ... co-occurrence degree calculation unit, 109 ... task model generation unit, 110 ... storage unit, 111 ... subdomain search pattern Generation unit 112 ... Web search unit 113 ... Search pattern matching unit 115 ... Related task search pattern generation unit 117 ... Co-occurrence degree calculation unit 118 ... Association processing unit 200 ... Search engine 300 ... Mobile terminal 300 ... Search engine.

Claims (12)

An input means for receiving input of a character string from the user;
A domain task search pattern generation means for generating a search pattern including an input character string received by the input means;
Domain task search means for searching for a web page including a search pattern generated by the domain task search pattern generation means;
From the Web page searched by the domain task search means, a search pattern character string indicated by the search pattern and a series of character strings including an operation character string representing an operation in connection with the search pattern character string are used as a domain task. A domain task extracting means for extracting;
A task model generation apparatus comprising: a storage unit that stores the domain task extracted by the domain task extraction unit and the access destination information of the Web page including the domain task in association with each other.   The domain task search pattern generation unit generates a search pattern by connecting a particle that is connected to an input character string received by the input unit to be connected to a word representing an action after the input character string. The task model generation device according to claim 1.   3. The task model generation apparatus according to claim 2, wherein the domain task search pattern generation unit generates a search pattern by connecting the particle "" to the end of the input character string.   The domain task extraction unit performs dependency analysis to determine an operation character string corresponding to the input character string, and extracts a series of character strings including the search pattern and the operation character string. The task model generation device according to any one of claims 1 to 3.   2. The task model generation according to claim 1, wherein the domain task extraction unit extracts, as a domain task, a character string that is connected immediately after the search pattern and includes only kanji representing an action together with the search pattern. apparatus. An inclusion task in which a concatenated character string for concatenating a domain task character string, which is a domain task extracted by the domain task extracting means, with a modifier for modifying the operation indicated by the domain task character string is added. Inclusion task search pattern generation means for generating a search pattern;
Further comprising
The domain task search means searches for a web page including the inclusion task search pattern generated by the inclusion task search pattern generation means,
The task according to any one of claims 1 to 5, wherein the domain task extraction unit extracts an inclusion task character string in which a modification character string indicating a modification word is connected to the inclusion task search pattern. Model generator. A domain co-occurrence degree calculating means for calculating a co-occurrence degree indicating a degree of association between the input character string and the action character string;
7. The domain task extraction unit extracts a domain task having a co-occurrence degree equal to or greater than a predetermined value based on a calculation result by the domain co-occurrence degree calculation unit. The task model generation device described in 1. An inclusion task co-occurrence degree calculating means for calculating a co-occurrence degree indicating a degree of association between the domain task and the modifier character string;
8. The domain task extraction unit extracts an inclusion task having a co-occurrence degree equal to or greater than a predetermined value based on a calculation result by the inclusion task co-occurrence degree calculation unit. The task model generation device according to item. An input means for receiving input of a character string from the user;
Sub-domain search pattern generation means for generating a search pattern including an input character string received by the input means;
Subdomain search means for searching for a web page including a search pattern generated by the subdomain search pattern generation means;
Subdomain extraction means for extracting, as a subdomain, a series of character strings composed of noun character strings linked to the search pattern from the web page searched by the domain task search means;
Related task search pattern generation means for generating a search pattern of related tasks from a subdomain character string indicating a subdomain extracted by the subdomain extraction means;
Related task search means for searching for a Web page including a search pattern generated by the related task search pattern generation means;
Related task extraction means for extracting, as a related task, a series of character strings composed of the search pattern character string and an action character string representing an action linked to the search pattern character string from the Web page searched by the related task search means When,
A task model generation device comprising: a storage unit that stores the related task extracted by the related task extraction unit and the access destination information of the Web page including the related task in association with each other. A domain task search pattern generation means for generating a search pattern including an input character string received by the input means;
Domain task search means for searching for a web page including a search pattern generated by the domain task search pattern generation means;
From the Web page searched by the domain task search means, a search pattern character string indicated by the search pattern and a series of character strings including an operation character string representing an operation in connection with the search pattern character string are used as a domain task. A domain task extracting means for extracting;
The storage means is a task model generation device that stores the domain task extracted by the domain task extraction means and the access destination information of the Web page including the domain task as a task model in association with each other,
further,
A related task co-occurrence degree calculating means for calculating a co-occurrence degree indicating a degree of association between a domain task and a related task made of a predetermined character string;
The task model generation device according to claim 9, further comprising an association processing unit that associates a related task with a domain task based on the co-occurrence degree calculated by the related task co-occurrence degree calculating unit. An input step for accepting input of a character string from the user;
A domain task search pattern generation step for generating a search pattern including the input character string received by the input step;
A search step of searching for a web page including the search pattern generated by the domain task search pattern generation step;
From the Web page searched by the search step, a search pattern character string indicated by the search pattern and a series of character strings including an action character string representing an action linked to the search pattern character string are extracted as a domain task. A domain task extraction step;
A storage step of storing the domain task extracted in the domain task extraction step and the access destination information of the Web page including the domain task in a storage unit in association with each other. An input step for accepting input of a character string from the user;
A subdomain search pattern generation step of generating a search pattern including the input character string received by the input step;
A subdomain search step of searching for a Web page including the search pattern generated by the subdomain search pattern generation step;
A subdomain extraction step of extracting a series of character strings consisting of noun character strings linked to the search pattern as subdomains from the web page searched by the subdomain search step;
A related task search pattern generation step of generating a subdomain character string indicating the subdomain extracted by the subdomain extraction step as a search pattern of the related task;
A related task search step of searching for a Web page including the search pattern generated by the related task search pattern generation step;
Related task extraction step for extracting a series of character strings consisting of the search pattern character string and an action character string representing an action linked to the search pattern character string as a related task from the Web page searched in the related task search step When,
And a storage step of storing the related task extracted in the related task extraction step and the access destination information of the Web page including the related task in a storage unit in association with each other.

Images