WO2016135905A1 - 情報処理システム及び情報処理方法 - Google Patents
情報処理システム及び情報処理方法 Download PDFInfo
- Publication number
- WO2016135905A1 WO2016135905A1 PCT/JP2015/055481 JP2015055481W WO2016135905A1 WO 2016135905 A1 WO2016135905 A1 WO 2016135905A1 JP 2015055481 W JP2015055481 W JP 2015055481W WO 2016135905 A1 WO2016135905 A1 WO 2016135905A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- evaluation
- expression
- evaluation expression
- data
- information processing
- Prior art date
Links
- 230000010365 information processing Effects 0.000 title claims abstract 30
- 238000003672 processing method Methods 0.000 title claims 3
- 238000011156 evaluation Methods 0.000 claims abstract description 333
- 230000014509 gene expression Effects 0.000 claims abstract description 316
- 238000003860 storage Methods 0.000 claims abstract description 21
- 238000000034 method Methods 0.000 claims description 36
- 238000000605 extraction Methods 0.000 claims description 20
- 238000013480 data collection Methods 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000004458 analytical method Methods 0.000 claims description 14
- 239000000284 extract Substances 0.000 claims description 5
- 238000012545 processing Methods 0.000 description 36
- 230000004044 response Effects 0.000 description 34
- 239000013598 vector Substances 0.000 description 28
- 230000008569 process Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 238000004891 communication Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 238000012937 correction Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 230000004308 accommodation Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 239000013604 expression vector Substances 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 235000012054 meals Nutrition 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000002040 relaxant effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
- G06F16/24578—Query processing with adaptation to user needs using ranking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2455—Query execution
- G06F16/24553—Query execution of query operations
- G06F16/24554—Unary operations; Data partitioning operations
- G06F16/24556—Aggregation; Duplicate elimination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/248—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/194—Calculation of difference between files
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/247—Thesauruses; Synonyms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/1822—Parsing for meaning understanding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
Definitions
- the present invention relates to a technique for presenting information to a user.
- the user is requested to input a keyword such as a word or a phrase, and a document, a Web page, a part of a sentence, a photograph, sound, or product information that is closely related to the keyword is presented.
- Information retrieval methods are widely used to extract information required by users from a large amount of media information such as documents or images. Further, there are techniques such as a similar search and an associative search in which not only a keyword input by a user but also information including a synonym and a word closely related thereto are searched.
- the estimation and presentation method is widely used as a recommendation technique, and in particular, a technique such as collaborative filtering is used.
- one aspect of the present invention is a memory that stores evaluation data that associates each of a plurality of objects with a plurality of evaluation expressions, and evaluation expression relation data that indicates a relationship between the evaluation expressions.
- a question generation unit that generates and outputs a question based on the evaluation data and the evaluation expression relation data, and when an answer to the question is input, the target included in the evaluation data based on the answer
- a matching unit that outputs the above information.
- the information presentation system is a system that narrows down and presents information requested by a user based on a question response with the user.
- the search target information is a document, an image, sound, or other data.
- FIG. 2 is a block diagram showing an example of a computer constituting the information presentation system of this embodiment.
- the computer 201 constituting the information presentation system of the present embodiment includes an input device 202, a display device 203, a communication device 204, a computing device (CPU) 205, and an external storage device 206.
- the input device 202 is a keyboard and mouse for inputting commands and the like.
- the input device 202 is a device for inputting a command executed for control of a program executed by the arithmetic unit (CPU) 205 and other control of connected devices.
- CPU arithmetic unit
- the display device 203 is a device such as a display that appropriately displays processing contents.
- the communication device 204 is used for exchanging data from an external device such as a PC or a server. Specifically, the communication device 204 is used for purposes such as acquisition of an execution command by a user from an external device and acquisition of information such as an image or text from an external device. The communication device 204 is also used for the purpose of transmitting the processing content in the computer 201 to an external device.
- an external device such as a PC or a server.
- the communication device 204 is used for purposes such as acquisition of an execution command by a user from an external device and acquisition of information such as an image or text from an external device.
- the communication device 204 is also used for the purpose of transmitting the processing content in the computer 201 to an external device.
- the arithmetic unit (CPU) 205 is an arithmetic unit that executes processing such as question answering with the user.
- the external storage device 206 is an external storage device such as an HDD or a memory.
- the external storage device 207 stores data necessary for answering questions and data to be searched.
- the external storage device 206 is also used to temporarily store data generated during processing executed by the arithmetic unit (CPU) 205.
- the computer 201 may not include the input device 202, the display device 203, and the communication device 204.
- a command or the like is input from an external device using the communication device 204.
- the processing result is transmitted to an external device using the communication device 204.
- the output and input of a module that executes processing may be performed via the external storage device 206. That is, when the processing unit 1 (not shown) outputs the processing result to the processing unit 2 (not shown) and the processing unit 2 receives the processing result as an input, the processing unit 1 actually stores the processing result in the external storage.
- the data may be output and stored in the device 206, and the processing unit 2 may acquire the output result of the processing unit 1 stored in the external storage device 206 as an input.
- FIG. 1 is a functional block diagram showing an example of the information presentation system of the present embodiment.
- the information presentation system of the present embodiment includes a question knowledge database generation device 101, a database 107, a question answering system 112, and a database 116.
- the question knowledge database generation apparatus 101 includes a data collection unit 102, an evaluation expression extraction unit 103, and an expression map generation unit 104.
- the database 107 includes a collection database (DB) 108, an individual evaluation database (DB) 109, an evaluation expression map 110, and a domain knowledge database (DB) 111.
- the question answering system 112 includes a question answer generating unit 113, a user answer acquiring unit 115, and a matching unit 114.
- the database 116 includes an individual evaluation DB 109, an evaluation expression map 110, and a domain knowledge DB 111 similar to the database 107.
- the information presentation system in FIG. 1 is realized by one or more computers 201.
- the question knowledge database generation apparatus 101 and the database 107 may be realized by a single computer 201, and the question answering system 112 and the database 116 may be realized by another computer connected thereto via a network.
- the data collection unit 102, the evaluation expression extraction unit 103, the expression map generation unit 104, the question answer generation unit 113, the user answer acquisition unit 115, and the matching unit 114 are stored in the external storage device 206 of each computer 201.
- the program is realized by the execution of the arithmetic unit 205, and the databases 107 and 116 are stored in the external storage device 206 of each computer 201.
- the information presentation system in FIG. 1 may be realized by one or more computers 201.
- each unit of the question knowledge database generation device 101 and the question answering system 112 is realized by the arithmetic device 205 executing a program stored in the external storage device 206 of one computer 201.
- the database 116 can be omitted by the question answering system 112 referring to the database 107.
- the database 107 is stored in the external storage device 206 of the computer 201 different from the question knowledge database generation device 101 and the question answering system 112, and the question knowledge database generation device 101 and the question answering system 112 are stored in the database via the network. 107 may be created and referenced.
- the configuration of the information presentation system of the present embodiment is not limited to the above example. That is, an arbitrary part of the information presentation system of the present embodiment may be realized by the computer 201 connected to the network, or realized by a virtual computer generated by logically dividing one computer 201. May be.
- the domain knowledge DB 111 is a database created in advance, and includes information related to a subject that is a theme (topic).
- the domain knowledge DB 111 will be described.
- the domain knowledge DB 111 may include, for example, information on the ontology of concepts related to travel.
- Such information includes, for example, information on is-a relationship, part-of relationship, instance-of relationship, and the like.
- hotel is an accommodation facility
- hotel is-a accommodation facility means that the hot spring is part of the hotel facilities, “hot spring part-of hotel”
- hotel A is a concrete concept of the hotel.
- Such an instance is expressed as “Hotel Ainstance-of Hotel”.
- Information arrangement methods other than the ontology format may be used.
- Prior knowledge about the theme is prepared in advance as a domain knowledge DB 111 by using manual or computer processing.
- the domain knowledge DB 111 may also include data related to rules for the data collection unit 102 to extract and classify documents related to the theme.
- the data collection unit 102 collects document data to be processed (for example, a website, a questionnaire, or any other type of document). For example, when presenting a recommended travel destination, the data collection unit 102 collects commercial facilities such as hotels, tourist facilities, and transportation facilities, home pages of public facilities, word-of-mouth, local information, blogs, and the like. When there is customer questionnaire information, the data collection unit 102 also collects such information. When presenting a product, the data collection unit 102 collects sites, documents, and the like related to the product. The collected information is classified for each type and stored in the collection DB 108.
- a processing example of the data collection unit 102 will be described.
- data is collected from the Web on the assumption that the travel destination is presented to the user on the theme of travel will be described.
- FIG. 3 is a flowchart illustrating an example of processing executed by the data collection unit 102 according to the present exemplary embodiment.
- the data collection unit 102 collects information by crawling the Web via a network connected to the communication device 204, for example.
- the domain knowledge DB 111 may be used.
- the characteristics of the pages to be collected are stored in the domain knowledge DB 111 in advance, and the data collection unit 102 collects the pages based on the features. It is conceivable that the domain knowledge DB 111 holds, for example, features such as a site that contains many keywords related to travel and a site that has information such as business hours and fees indicating a commercial facility.
- the data collection unit 102 adds tags indicating types to those sites. For example, a tag indicating the type of facility such as a hotel, a hot spring, a department store, a tag indicating a location, a tag indicating a distinction of information sources such as a word of mouth, a blog, and an owner site can be considered. Tagging rules are also stored in the domain knowledge DB 111. By attaching tags in this way, it is possible to determine page reliability, objectivity, and the like, for example, word-of-mouth information can obtain more objective evaluation information than owner page. The reliability of the page may be digitized and stored as an attribute.
- the data collection unit 102 may collect data not only from the Web but also from documents in the organization and tag them. Data collected and tagged by the data collection unit 102 is stored in the collection DB 108.
- the evaluation expression extraction unit 103 extracts, from the collected data, an evaluation expression such as “Hotel A hot spring has a good view and relaxation”, the expression to be evaluated, an evaluation expression for the object, and If possible, estimate the evaluator's attributes.
- the method of generating the collection DB 108 as described above is an example, and the processing of this embodiment described later can be executed using the collection DB 108 generated by a method other than the above.
- FIG. 4 is a flowchart illustrating an example of processing executed by the evaluation expression extraction unit 103 according to this embodiment.
- the process of the evaluation expression extraction unit 103 will be described with reference to FIG.
- the domain knowledge DB 111 is referred to as necessary, but is omitted in FIG.
- the evaluation expression extraction unit 103 analyzes the layout of the page (document) stored in the collection DB. It also analyzes the meaning of each part of the page. An example of this analysis will be described with reference to FIG.
- FIG. 5 is an explanatory diagram of an example of data collected by the data collection unit 102 of this embodiment.
- a page 501 illustrated in FIG. 5 is an example of a page such as a word-of-mouth website of a hotel.
- the upper part 502 of the page 501 includes the hotel name, and the lower part 503 includes the evaluation score of the entire word of mouth or an explanatory text about the hotel.
- the lower part 504 displays the hotel's evaluation (so-called word-of-mouth) written by the individual user of the hotel.
- An evaluation score as shown in the portion 503 may be indicated for each word of mouth.
- There are various layouts such as an advertisement or related facility information displayed on the right or bottom of the page 501.
- the evaluation expression extraction unit 103 analyzes the layout of the page 501, and estimates the meaning of each unit (the part 502 is the title, the part 504 is the word of mouth, the part 503 is the description and the evaluation score, etc.). Rules and keywords for estimating the layout are stored in the domain knowledge DB 111 in advance, and the evaluation expression extraction unit 103 refers to them. For example, it can be determined that the portion 502 is a title because it is at the top of the page and the font is large. In addition, it is judged that the title is a hotel name because a keyword representing a concept related to a hotel often appears in the body of the page even if the word “hotel A” is not included, such as “hotel A”. be able to. These concepts are held in the domain knowledge DB 111.
- the layout structure can be extracted by using an HTML tag or the like.
- a layout analysis technique used for document processing may be used for the layout analysis step 401.
- the evaluation expression extraction unit 103 extracts a part in which the evaluation text is described as a result of the layout analysis.
- the evaluation text is described in the portion 503 or 504.
- the evaluation sentence is a sentence including an expression for evaluating something such as “XX is beautiful”. However, there are cases where the evaluation target (in this example, “XX”) is not specified. Since the evaluation is often expressed by an adjective, for example, a list of adjectives used as the evaluation expression is stored in the domain knowledge DB 111 in advance, and the evaluation expression is extracted by extracting a sentence including any of the adjectives. Can be extracted.
- the characteristics (rules) of the evaluation expression based on the syntax analysis result may be determined.
- a polarity analysis technique may be used.
- the evaluation expression extraction unit 103 extracts the evaluation expression from the sentences extracted in the evaluation sentence extraction step 402. For example, in the example of the sentence “Hotel C has a good meal and a large bed so you can relax. There is an open-air bath, which is recommended.” “Delicious” “Wide” “Relaxing” “Recommended” It becomes evaluation expression.
- the evaluation expression extracting unit 103 extracts the evaluation object of the evaluation expression extracted in the evaluation expression extracting step 403.
- the evaluation expression extraction unit 103 estimates the evaluation target from the peripheral information. For example, in the case of a word-of-mouth site, the facility name or product name to be evaluated may be described in the title or the like, so the evaluation object is estimated from the title or the like.
- the evaluation expression extraction unit 103 estimates the evaluator's attributes from the text.
- the attribute of the evaluator is, for example, a situation surrounding the evaluator (premise of evaluation) that can be read from the sentence such as, for example, traveling in a family or using a car when traveling.
- the degree of specialization can be determined by analyzing characteristics such as a highly specialized keyword appearing in a sentence and a high degree of concreteness in expression.
- the evaluator attribute estimation step 405 may be omitted. In particular, it may be omitted for documents that are difficult for the evaluator to estimate.
- individual evaluation DB 109 individual objects such as a hotel A and a hotel B and evaluation expressions for the objects are arranged and stored. If evaluator attributes are also estimated, they are organized as a database.
- FIG. 6 is an explanatory diagram of an example of data included in the individual evaluation DB 109 of the present embodiment.
- the fluctuation may be corrected and the normalized evaluation expression may be stored in the individual evaluation DB 109.
- “delicious”, “delicious”, “delicious”, and the like may be collected in the evaluation expression “delicious” and recorded in the individual evaluation DB 109.
- the data 601 in FIG. 6 includes information about individual targets such as an individual target ID 601A, its target type 601B, and a name 601C. Although illustration is omitted, in addition, the acquired information such as the Web page address and the nearest station can be included in the data 601 and stored in the individual evaluation DB 109.
- the 6 includes an evaluation expression 602B for each object identified by the object ID 602A.
- the same object ID and the same evaluation expression may appear multiple times.
- the ID 602C of the evaluator type can be included in the data 602 and stored in the individual evaluation DB 109.
- the evaluator type ID 602C may be omitted.
- the evaluator type for the object and the evaluation expression A plurality of values may be included as the ID 602C.
- the data 602 may further include a tag indicating the type of document from which each evaluation expression is extracted.
- tags indicating a plurality of types are associated with the evaluation expression.
- the 6 includes an object ID 603B associated therewith for each object ID 603A.
- the hotel A identified by the target ID “0012327” and the restaurant B identified by the target ID “083181” are both targets included in the data 601, but the restaurant B is attached to the hotel A. Therefore, “083181” is held as the associated target ID 603B corresponding to the target ID “0012327”.
- the associated object ID 603B corresponding to the object having no associated object is empty, and when a plurality of objects are associated with one object, the IDs of the plurality of objects are retained as the associated object ID 603B.
- the evaluator type 6 includes an evaluator type ID 604A and an attribute 604B of the evaluator type.
- the evaluator type may appear multiple times.
- the attributes include those represented by numerical values.
- the expression map generation unit 104 analyzes the relationship between the evaluation expressions used for each target type.
- the evaluation expression may differ in expression used for each object and its meaning. For example, the evaluation expression “easy” for ramen is rarely used for facilities such as hotels. In addition, the meaning is different from the evaluation that people are “light”. Therefore, it is necessary to analyze what evaluation expression is used and how it is used for each target type. In this example, relationships such as similarity relationships, inclusion relationships, and trade-off relationships between evaluation expressions are analyzed.
- FIG. 7 is a flowchart illustrating an example of processing executed by the expression map generation unit 104 of the present embodiment.
- the expression map generation unit 104 collects evaluation expressions used for the evaluation for each target type, and analyzes relationships such as similarity relationships, inclusion relationships, and trade-off relationships between the expressions.
- the expression map generation unit 104 first collects evaluation expressions for each object for each object type.
- FIG. 8 is an explanatory diagram of an example of evaluation expressions collected by the expression map generation unit 104 of the present embodiment.
- the evaluation expression 801 in FIG. 8 is a collection of evaluation expressions for each object for the object type “hotel”. Each row lists one or more evaluation expressions 801B for one object having the name 801A.
- the expression map generation unit 104 first quantifies the proximity between evaluation expressions.
- an example of the method will be described. Assume that there are two evaluation expressions E1 and E2. At this time, the expression map generation unit 104 calculates a distance D (E1, E2) between E1 and E2.
- D the number of elements in the set S is expressed as #
- the expression map generation unit 104 determines that the evaluation expression 801B of the same target (for example, hotel A) co-occurs when both evaluation expressions E1 and E2 are included, and does not co-occur when only one is included. Then, it is determined whether or not the evaluation expressions E1 and E2 co-occur for all the objects corresponding to the object type “hotel”, and the frequency of co-occurrence is calculated from the result of the determination.
- the distance between evaluation expressions can be defined by, for example, Expression (1).
- D (E1, E2) Log [(2 ⁇ #
- D (E1, E2) is 0 when both E1 and E2 always appear (that is, co-occur) in an object in which at least one of the evaluation expressions E1 or E2 appears, and is Log 2 when no co-occurs at all. Become. Further, let F (K, E) be the number of times the evaluation expression E appears as the evaluation expression of the target K. At this time, it can be estimated that E1 and E2 are closer when the number of times F (K, E1) and F (K, E2) at which E1 and E2 appear is closer. As another example of the calculation of the distance D (E1, E2) Equation (2) may be used.
- D (E1, E2) ⁇ [(
- the distance between the evaluation expressions can be quantified as a function based on the co-occurrence frequency of the evaluation expressions E1 and E2, the closeness of the number of appearances of E1 and E2, and the like.
- the expression map generation unit 104 estimates such an inclusion relationship between expressions.
- the degree to which the evaluation expression E2 is included in the evaluation expression E1 is often the evaluation expression E1 if the evaluation expression E2 appears in the evaluation expression 801B of the target K (in other words, the evaluation expressions E1 and E2 co-occur).
- the expression map generation unit 104 calculates Expression (4) for the evaluation expression E (for example, each of E1 and E2).
- ⁇ is calculated for the entire target K of the target type being considered.
- the expression map generation unit 104 calculates P (K, E) by Expression (5).
- This P (K, E) can be regarded as a probability distribution of the evaluation expression E when K is considered as a variable. Therefore, the expression map generation unit 104 calculates, for example, the negative Cullback-Liblar information amount I (E1, E2) of P (K, E1) and P (K, E2) by Expression (6).
- the degree to which the evaluation expression E2 is included in the evaluation expression E1 can be quantified. This is a large value if E1 also appears with high probability when E2 appears.
- the expression map generation unit 104 may estimate the similarity relationship and the inclusion relationship between the evaluation expressions using the synonym dictionary and the dictionary indicating the inclusion relationship.
- the expression map generation unit 104 quantifies the trade-off relationship between the evaluation expressions. For example, “has a sense of luxury” and “cheap” tend to be in a trade-off relationship. Such a relationship can be acquired from an evaluation expression using an antonym dictionary, and from a linguistic expression representing a conflicting relationship such as “high-quality but cheap”. In addition, evaluation expressions that are in a trade-off relationship may be difficult to co-occur, and the reciprocal number T (E1, E2) of D (E1, E2) defined above may be calculated by equation (7). .
- the expression map generation unit 104 analyzes the similarity relationship, the inclusion relationship, and the trade-off relationship between evaluation expressions.
- the analysis result is stored in the evaluation expression map 110.
- information such as the degree of similarity D (E1, E2) between expressions, the degree of inclusion I (E1, E2), and the degree of trade-off T (E1, E2) is stored in the evaluation expression map 110.
- information such as an antonym may be stored as information indicating a trade-off relationship.
- the expression map generation unit 104 shares the relationship analysis result between the evaluation expressions between the similar object types.
- the processing in the inter-expression relationship estimation step 701 is performed for each target type. However, since the similar target types, for example, “hotel” and “hotel” are similar, it can be expected that the relationship analysis results between the evaluation expressions related to each other can be used for each other.
- the sum ( ⁇ ) is calculated for all evaluation expressions E.
- P (O1, E) can be regarded as a probability distribution when E is a random variable.
- the expression map generation unit 104 calculates a distance L (O1, O2) between the probability distributions P (O1, E) and P (O2, E).
- a Cullback / liver distance can be used. Since object types with similar appearance frequencies of the evaluation expression E are considered to be similar, the similarity S (O1, O2) is defined, for example, as in Expression (9).
- the similarity can be measured based on how the evaluation expression is used and whether the usage frequency is similar between the target types O1 and O2.
- the expression map generation unit 104 can redefine the similarity (or distance), inclusion relation, and trade-off relation between evaluation expressions using the similarity between target types measured in this way. For example, it is assumed that the distance D (O1; E1, E2) between the evaluation expression E1 and the evaluation expression E2 in the target type O1 is defined by the method of the inter-expression relationship estimation step 701. At this time, the distance between the evaluation expressions in the target type O1 can be redefined as, for example, Expression (10).
- the sum ( ⁇ ) is calculated for all target types O. This shares the distance between evaluation expressions of other target types with the similarity S between the target types as a weight.
- the sum is calculated for all target types O.
- the sum may be calculated only for a predetermined target type or a target type having a certain degree of similarity or more. Similar processing may be applied to the inclusion relationship and the trade-off relationship.
- the expression map generation unit 104 does not have to perform the similarity estimation step 702 between the target types. If it cannot be obtained, the relationship between the evaluation expressions can be shared by performing this process.
- the relationship between two evaluation expressions was analyzed, but it is convenient to express the evaluation expression as a vector and express the evaluation expression as a point on the vector space. Therefore, the evaluation expression may be converted into a vector so that expressions with high similarity are arranged in the vector space.
- a method such as Force-Directed Algorithm can be used. In this method, a constant repulsive force is defined between all elements, an attractive force acting between elements is defined based on the similarity between the elements, and the energy of the entire system based on the attractive force and the repulsive force is reduced. The arrangement of elements is sequentially corrected, and when the arrangement converges, the position of the element is determined.
- the expression map generation unit 104 converts the evaluation expression into a vector as described above, and stores the obtained vector space and vector value in the evaluation expression map 110 as a similarity map.
- vector spaces can be defined for inclusion relations and trade-off relations.
- the question answering system 112 hears the user's preference based on the question to the user and the user's response to the question, narrows down the target close to the user's preference by matching, and presents it to the user.
- the question answering system 112 uses the individual evaluation DB 109 and the evaluation expression map 110 generated by the question knowledge database generation device 101.
- the question response generation unit 113 generates a question using the evaluation expression, and estimates the user's preference by repeating the process of obtaining an answer from the user.
- FIG. 9 is a flowchart showing an example of processing executed by the question answering system 112 of this embodiment.
- Steps 901 to 904 are processing of the question answer generation unit 113
- step 905 is processing of the user answer acquisition unit 115
- steps 906 to 909 are processing of the matching unit 114.
- the question response generation unit 113 generates a question evaluation expression candidate list for each target type.
- FIG. 10A is an explanatory diagram of an example of an evaluation expression candidate list for questions generated by the question response generation unit 113 of the present embodiment.
- the evaluation expression candidate list 1001 shown in FIG. 10A summarizes the number of appearances of evaluation expressions for each target.
- the number in parentheses below the evaluation expression indicates the number of appearances of the evaluation expression.
- the question response generation unit 113 removes expressions with a low appearance frequency as evaluation expressions, and creates a list of evaluation expressions in which the appearance frequency is a certain value or more.
- the question response generation unit 113 adds evaluations regarding the presence of facilities and equipment (for example, “there is an open-air bath”, “there is a restaurant”, etc.) to the evaluation expression candidate list 1001.
- Such information can be acquired from a target owner page such as a hotel stored in the collection DB 108, for example.
- the question response generation unit 113 generates a question such as “Is the hotel close to the station?” Based on the evaluation expression.
- the user answer acquisition unit 115 accepts a reply in a natural language sentence from the user, or selects an answer (for example, “prefer close”, “preferably”, “do not care too much”, “prefer distant if possible”, “distant The user's selection is accepted as an answer, and the matching unit 14 narrows down the target that matches the user's preference based on the accepted answer.
- the question answer generation unit 113 presents an efficient question order.
- An example of processing will be described.
- the question response generation unit 113 leaves an expression estimated to apply to the target with high accuracy among the evaluation expressions for each target in the evaluation expression candidate list 1001, and excludes the rest.
- the high degree of accuracy is based on, for example, the high frequency of occurrence of the corresponding evaluation expression and the low frequency of appearance of the evaluation expression that is in opposition to the evaluation expression (ie, the degree of trade-off relationship is large). Can be calculated.
- the question response generation unit 113 may determine that the accuracy of the evaluation expression is lower as the co-occurrence frequency of a certain evaluation expression and the evaluation expression opposed to the evaluation expression is higher. In addition, since the existence of facilities and equipment can be acquired from an owner page or the like, it is considered that the accuracy is high. As described above, the question response generation unit 113 uses, for each evaluation expression, a predetermined accuracy (for example, whether it is an owner page or an individual blog) according to the type of document extracted. Alternatively, a weight of accuracy) may be given. Therefore, the question response generation unit 113 creates information indicating whether or not each target evaluation is applicable to each evaluation expression.
- a predetermined accuracy for example, whether it is an owner page or an individual blog
- a weight of accuracy may be given. Therefore, the question response generation unit 113 creates information indicating whether or not each target evaluation is applicable to each evaluation expression.
- FIG. 10B is an explanatory diagram of an example of information generated by the question response generation unit 113 according to the present embodiment and indicating whether or not each target evaluation applies to each evaluation expression.
- a table 1002 indicating whether or not the evaluation of each target is applicable to each evaluation expression indicates that, for example, it is determined that the probability that the hotel A is “close to the station” is high. Further, the table 1002 indicates that the accuracy of the evaluation “relaxed” is low or absent for the hotel A.
- the question answer generation unit 113 repeats the question to the user, and is estimated to be efficient when narrowing the number of candidates that match the user's answer to a predetermined threshold value or less. Calculate the order of questions. For example, in order to calculate an efficient question order, the question response generation unit 113 assumes that the answers to the question based on the evaluation expression as described above are only “Yes” and “No”. Even if an answer is obtained, a question order is generated so that questions that can be excluded from many candidates are output preferentially.
- the question response generation unit 113 ranks the questions so that questions that are considered to be more efficient are output earlier, and calculates an evaluation expression that is the basis of a predetermined number of upper questions. Keep it. The value used for this ranking (here, the total number of candidates) is used as the score.
- the question response generation unit 113 estimates the number of objects that do not correspond to the answer to the question based on each evaluation expression (that is, is excluded by the answer), and based on the estimated number of objects (for example, A score is calculated (so that a question based on an evaluation expression whose number satisfies a predetermined condition is output earlier). For example, the question response generation unit 113 counts all the objects, the number of objects corresponding to the evaluation expression “close to the station”, and an evaluation expression in a trade-off relationship with the evaluation expression (for example, “far from the station”). The score may be calculated based on the relationship between the number of objects corresponding to).
- the ratio of the number of objects that do not correspond to either of the evaluation expressions “close to the station” or “far from the station” to the total number of objects is large. Even if the answer to the question is “Yes” or “No”, it indicates that the ratio of targets that cannot be excluded from candidates is large. For example, when one of the number of objects corresponding to the evaluation expression “close to the station” and the number of objects corresponding to “far from the station” is extremely small, the answer is “Yes” or “No”. In the case of one of the above, there are few targets that can be excluded from the candidates, and in the case of the other, there are almost no targets that remain as candidates.
- the question response generation unit 113 for example, a question based on an evaluation expression in which the ratio of an object that does not correspond to either the evaluation expression or the expression opposite to the evaluation expression is larger than a predetermined value, and an object corresponding to one answer
- the score of each evaluation expression may be calculated so that questions that do not correspond to the question based on the evaluation expression whose number is smaller than a predetermined value are output earlier. As a result, the target can be narrowed down efficiently.
- the question response generation unit 113 In the question order correction rule calculation step 903, the question response generation unit 113 generates a question order that considers not only the efficiency of narrowing down but also the naturalness of dialogue. In conversation, it is more natural to ask questions in a broad (eg, abstract or vague) expression at first, and then listen to specific details little by little, rather than listening to specific things from the beginning. In some cases, the user's mind changes during the conversation, or the user makes a wrong answer. Furthermore, there are cases where the user's needs may require a trade-off relationship, such as seeking a “cheap hotel” and a “hotel with a large room”, so compromises somewhere. , Etc. need to be adjusted.
- the question response generation unit 113 determines the priority order of the questions obtained in the efficient question order calculation step 902 based on the similarity relationship, the inclusion relationship, and the trade-off relationship between the evaluation expressions. Correct. At this time, the question response generation unit 113 uses the evaluation expression map 110.
- the question response generation unit 113 holds such vectors, and for the evaluation expressions E1 and E2 obtained in the efficient question order calculation step 902, the rank of E2 is lower than the rank of E1, and E2 includes E1. If E2 has not been used for the question yet, the score of E2 is increased according to the degree of inclusion. As a result, questions with a high degree of abstraction that have not been asked yet tend to be placed higher. As a result, in the above example, the question based on the evaluation expression E2 is likely to be output earlier than the question based on the evaluation expression E1.
- the question response generation unit 113 has already You may ask a question using an evaluation expression similar to the question made. This assumes that the user's mind changes and asks a similar question. Therefore, even in the case of evaluation expressions not listed in the efficient question order calculation step 902, the question order of evaluation expressions similar to the evaluation expressions that appeared in the past questions is advanced according to a certain rule (for example, randomly) ( For example, the highest position at that time).
- a certain rule for example, randomly
- the question response generation unit 113 makes a question using the evaluation expression at the top when the question order correction rule calculation step 903 is completed. For example, when the evaluation expression “close to the station” for the object “hotel” is selected as the highest-level evaluation expression, the question answer generation unit 113 may ask, “Is the hotel better near the station?” Output a question.
- the target type has been described as being fixed with an example of a hotel, but the question answer generation unit 113 includes a question evaluation expression candidate generation step 901, an efficient question order calculation step 902, and The processing of the question order correction rule calculation step 903 is performed on a plurality of target types in parallel, and the question selection step 904 selects a target type from among these according to a certain rule, and generates a question. Good.
- the user response acquisition unit 115 acquires a response from the user.
- the user answer acquisition unit 115 may acquire an answer described in a natural language, or may prepare a plurality of default answers and acquire a user's selection from them as an answer.
- the matching unit 114 expresses the answer result by quantifying it. For example, it may be expressed by the answer result vector described above.
- the matching unit 114 selects a candidate that matches the user's answer result.
- An example of calculating the degree of coincidence with the user's answer result will be described.
- the matching unit 114 expresses each object by a vector corresponding to each evaluation expression in the same manner as the answer result vector, and sets 1 if the evaluation expression exists, and 0 otherwise. .
- w1 (w1,..., Wn)
- the accuracy of each evaluation item corresponding to each object shown in FIG. 10B may be an element of an evaluation expression vector corresponding to each object.
- the value of the element corresponding to the evaluation expression “close to the station” is “1”
- the value of the element corresponding to the evaluation expression “relaxed” is “ 0 ".
- the matching unit 114 calculates the similarity between the vector generated from the evaluation expression corresponding to each object and the answer result vector, and determines that an object having a certain degree of similarity matches the answer result of the user. To do.
- the similarity between vectors For example, cosine similarity can be used. In this manner, for each target type, a target that matches the user's answer result can be selected as described above.
- the matching unit 114 needs to narrow down to compatible candidates with different target types. For example, in the case of presenting a travel plan, when there are two target types “hotel” and “sightseeing place (location)”, the location of the hotel A that matches the user's answer result and the user's answer Sightseeing locations that match the results should be the same or close. Accordingly, the matching unit 114 creates a combination of compatible candidates as a plan from among the candidates selected for each target type, and calculates them as candidates.
- the value of the answer result vector is given a value only to the element corresponding to the evaluation expression from which the answer is obtained from the user. If there is a value for similar evaluation expressions, both may share the answer according to the degree of similarity. For example, when the element value of an evaluation expression E ′ similar to a certain evaluation expression E is v and their similarity is s (assuming that s is normalized to be between 0 and 1), the evaluation The element value of the expression E may be defined as v * s.
- the matching unit 114 matches the user's answer in consideration of the attribute.
- the target may be selected. Specifically, for example, when an evaluator's attribute that matches the responding user is known, the matching unit 114 gives a higher weight to the appearance frequency of the evaluation expression used by the evaluator of the attribute. In this way, the appearance frequency of each evaluation expression is weighted, and a vector of evaluation expressions corresponding to each target is generated based on the appearance frequency of the weighted evaluation expression, and the similarity between it and the answer result vector is calculated. May be.
- the attribute of the evaluator suitable for the user may be specified by the user himself (for example, the user wants to emphasize the evaluation of a highly specialized person or the person who traveled with a family), or the user inputs
- the question answering system 112 may estimate based on the answered answer. As a result, it is possible to present an object more suitable for the user.
- the matching unit 114 performs the same process as the process related to the evaluator attribute for the type. be able to.
- the matching unit 114 determines whether or not to end the question. If not completed, the process proceeds to an efficient question order calculation step 902 to generate a question again. In the case of ending, the process proceeds to information presentation step 909. For example, the matching unit 114 may determine whether there is a target candidate that matches the user's answer result based on a certain standard, and may determine that the question is to be terminated if the number is equal to or less than a predetermined number. .
- the matching unit 114 presents candidate candidates that match the user's answer via the display device 203.
- the matching unit 114 may present a plurality of candidates or the one with the highest degree of match. Moreover, you may show for every object classification, for example, you may present the combination (plan) of object classification, such as the hotel A and the restaurant A.
- a travel plan or a tour in which a hotel, a transportation facility, a restaurant, etc. are set may be provided in advance.
- the question answering system 112 presents a plan or tour with a high degree of similarity by attaching an evaluation tag to the tour or plan in advance and measuring the degree of similarity between the evaluation tag and the user response result. Also good.
- the question answering system 112 estimates an evaluation for each evaluation expression from the action history, generates an action history vector in the same manner as the user answer result vector, and both vectors
- the information to be presented may be determined in consideration of the similarity to.
- facilities such as hotels and restaurants used for travel are shown as examples of search objects.
- other objects such as books, movies, any kind of retail goods, any kind
- the present invention can also be applied to searches for other facilities and real estate.
- each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor.
- Information such as programs, tables, and files that realize each function is a memory, hard disk drive, storage device such as SSD (Solid State Drive), or computer-readable non-transitory data such as an IC card, SD card, or DVD. It can be stored in a storage medium.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
Claims (15)
- 複数の対象の各々と複数の評価表現とを対応付ける評価データ、及び、前記評価表現間の関係を示す評価表現関係データを格納する記憶部と、
前記評価データ及び前記評価表現関係データに基づいて質問を生成して出力する質問生成部と、
前記質問に対する回答が入力されると、前記回答に基づいて前記評価データに含まれる前記対象の情報を出力するマッチング部と、を有することを特徴とする情報処理システム。 - 請求項1に記載の情報処理システムであって、
前記質問生成部は、
前記複数の評価表現に基づいて複数の質問を生成し、
記各評価表現に基づく質問に対する回答に対応しない前記対象の数を、前記評価データに基づいて推定し、前記推定した対象の数に基づいて前記複数の質問の出力順序を算出することを特徴とする情報処理システム。 - 請求項2に記載の情報処理システムであって、
前記評価表現関係データは、評価表現間のトレードオフ関係を示す情報を含み、
前記質問生成部は、前記各評価表現の出現頻度、及び、前記各評価表現とトレードオフ関係にある評価表現の共起頻度に基づいて、前記各評価表現が前記各対象に当てはまる確度を推定し、前記確度が高いと推定される複数の前記評価表現に基づいて前記複数の質問を生成することを特徴とする情報処理システム。 - 請求項2に記載の情報処理システムであって、
前記記憶部は、前記各評価表現が抽出された文書データの種類を特定する情報を格納し、
前記質問生成部は、特定された前記文書データの種類に基づいて前記各評価表現が前記各対象に当てはまる確度を推定し、前記確度が高いと推定される複数の前記評価表現に基づいて前記複数の質問を生成することを特徴とする情報処理システム。 - 請求項2に記載の情報処理システムであって、
前記評価表現関係データは、評価表現間の包含関係を示す情報を含み、
前記質問生成部は、前記複数の評価表現のうち第1評価表現が第2評価表現を包含する場合、前記第1評価表現に基づく質問を、前記第2評価表現に基づく質問より早く出力するように、前記出力順序を変更することを特徴とする情報処理システム。 - 請求項2に記載の情報処理システムであって、
前記マッチング部は、出力された一つ以上の前記質問に対する一つ以上の回答と、前記各対象に対応する評価表現との類似度を算出し、前記類似度が所定の条件を満たす前記対象の情報を出力することを特徴とする情報処理システム。 - 請求項6に記載の情報処理システムであって、
前記評価データは、前記各評価表現を使用した評価者の属性を示す情報を含み、
前記マッチング部は、前記各対象に対応する前記各評価表現の出現頻度に、前記各評価表現を使用した評価者の属性による重み付けをし、前記重み付けされた出現頻度に基づいて、出力された一つ以上の前記質問に対する一つ以上の回答と、前記各対象に対応する評価表現との類似度を算出することを特徴とする情報処理システム。 - 請求項6に記載の情報処理システムであって、
前記評価表現関係データは、評価表現間の類似関係を示す情報を含み、
前記質問生成部は、前記類似度が所定の条件を満たす前記対象の数が所定の条件を満たす場合、既に出力した質問の基となった評価表現に類似する評価表現に基づく質問を早く出力するように前記出力順序を変更することを特徴とする情報処理システム。 - 請求項1に記載の情報処理システムであって、
複数の文書データから前記複数の対象の各々に対応する前記評価表現を抽出して前記評価データを生成する評価表現抽出部と、前記複数の文書データから前記評価表現間の関係を推定して前記評価表現関係データを生成する評価表現関係データ生成部と、をさらに有することを特徴とする情報処理システム。 - 請求項9に記載の情報処理システムであって、
前記評価表現抽出部は、
レイアウト解析によって各文書データのうち前記対象が記述された部分及び前記評価表現が記述された部分を特定し、
前記評価表現が記述された部分から、品詞又は予め定められた規則に基づいて一つ以上の前記評価表現を抽出し、
前記対象が記述された部分から、前記抽出された評価表現に対応する前記対象を抽出し、
前記各文書データに含まれる文言に基づいて、前記各評価表現を使用した評価者の属性を推定し、
前記抽出した評価表現、前記抽出した対象及び前記推定した属性を含む前記評価データを前記記憶部に格納することを特徴とする情報処理システム。 - 請求項9に記載の情報処理システムであって、
前記評価表現関係データ生成部は、前記複数の文書データから、前記対象の種類ごとに、前記評価表現間の関係を推定して前記評価表現関係データを生成することを特徴とする情報処理システム。 - 請求項11に記載の情報処理システムであって、
前記評価表現関係データ生成部は、
二つの前記評価表現の共起頻度に基づいて前記二つの評価表現の類似関係を推定し、
二つの前記評価表現の共起頻度が所定の条件を満たす場合において、前記二つの評価表現が共起しないときの前記各評価表現の出現頻度の偏りに基づいて前記二つの評価表現の包含関係を推定することを特徴とする情報処理システム。 - 請求項12に記載の情報処理システムであって、
前記評価表現関係データ生成部は、前記対象の種類間の類似関係を推定し、
互いに類似すると推定された複数の種類の前記対象に対応する複数の前記評価表現に基づいて前記評価表現関係データを生成することを特徴とする情報処理システム。 - 請求項9に記載の情報処理システムであって、
前記記憶部には、収集されるべき前記文書データの特徴を示す情報及び前記文書データの種類を判別する規則を含む知識データがさらに格納され、
前記知識データに基づいて、ネットワークを介して前記複数の文書データを収集し、収集した前記複数の文書データの各々を前記文書データの種類を示す情報と対応付けて前記記憶部に格納するデータ収集部をさらに有することを特徴とする情報処理システム。 - 演算部と、前記演算部に接続される記憶部と、を有する計算機システムによる情報処理方法であって、
前記演算部には、複数の対象の各々と複数の評価表現とを対応付ける評価データ、及び、前記評価表現間の関係を示す評価表現関係データが格納され、
前記情報処理方法は、
前記演算部が前記評価データ及び前記評価表現関係データに基づいて質問を生成して出力する手順と、
前記質問に対する回答が入力されると、前記演算部が前記回答に基づいて前記評価データに含まれる前記対象の情報を出力する手順と、を含むことを特徴とする情報処理方法。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US15/551,747 US10671619B2 (en) | 2015-02-25 | 2015-02-25 | Information processing system and information processing method |
JP2017501757A JP6381775B2 (ja) | 2015-02-25 | 2015-02-25 | 情報処理システム及び情報処理方法 |
PCT/JP2015/055481 WO2016135905A1 (ja) | 2015-02-25 | 2015-02-25 | 情報処理システム及び情報処理方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2015/055481 WO2016135905A1 (ja) | 2015-02-25 | 2015-02-25 | 情報処理システム及び情報処理方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2016135905A1 true WO2016135905A1 (ja) | 2016-09-01 |
Family
ID=56788610
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/055481 WO2016135905A1 (ja) | 2015-02-25 | 2015-02-25 | 情報処理システム及び情報処理方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US10671619B2 (ja) |
JP (1) | JP6381775B2 (ja) |
WO (1) | WO2016135905A1 (ja) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6250121B1 (ja) * | 2016-09-16 | 2017-12-20 | ヤフー株式会社 | 地図検索装置、地図検索方法、および地図検索プログラム |
EP3327592A1 (en) * | 2016-11-25 | 2018-05-30 | Panasonic Intellectual Property Management Co., Ltd. | Information processing method, information processing apparatus, and non-transitory recording medium |
CN108153800A (zh) * | 2016-12-06 | 2018-06-12 | 松下知识产权经营株式会社 | 信息处理方法、信息处理装置以及程序 |
JP2018092582A (ja) * | 2016-12-06 | 2018-06-14 | パナソニックIpマネジメント株式会社 | 情報処理方法、情報処理装置、及び、プログラム |
JP2018173784A (ja) * | 2017-03-31 | 2018-11-08 | 富士通株式会社 | 検索処理方法、検索処理プログラム、および検索処理装置 |
JP2020140692A (ja) * | 2019-03-01 | 2020-09-03 | 楽天株式会社 | 文抽出システム、文抽出方法、及びプログラム |
JP2021039715A (ja) * | 2019-08-30 | 2021-03-11 | バイドゥ オンライン ネットワーク テクノロジー (ベイジン) カンパニー リミテッド | コンテンツ埋込方法、装置、電子デバイス、記憶媒体、及びプログラム |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102535044B1 (ko) * | 2015-12-08 | 2023-05-23 | 삼성전자주식회사 | 단말장치, 서버 및 이벤트 제안방법 |
CN108052577B (zh) * | 2017-12-08 | 2022-06-14 | 北京百度网讯科技有限公司 | 一种通用文本内容挖掘方法、装置、服务器及存储介质 |
US11238508B2 (en) * | 2018-08-22 | 2022-02-01 | Ebay Inc. | Conversational assistant using extracted guidance knowledge |
JP7200683B2 (ja) * | 2019-01-11 | 2023-01-10 | 富士フイルムビジネスイノベーション株式会社 | 情報処理装置及びプログラム |
US11036941B2 (en) * | 2019-03-25 | 2021-06-15 | International Business Machines Corporation | Generating a plurality of document plans to generate questions from source text |
JP7434125B2 (ja) * | 2020-09-16 | 2024-02-20 | 株式会社東芝 | 文書検索装置、文書検索方法、及びプログラム |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013089109A (ja) * | 2011-10-20 | 2013-05-13 | Nippon Telegr & Teleph Corp <Ntt> | 質問項目推薦装置、質問項目推薦方法及び質問項目推薦プログラム |
WO2013080406A1 (ja) * | 2011-11-28 | 2013-06-06 | Necソフト株式会社 | 対話システム、冗長メッセージ排除方法および冗長メッセージ排除プログラム |
JP2013214294A (ja) * | 2012-03-06 | 2013-10-17 | Okwave:Kk | クライアントシステム及びサーバ |
JP2014164351A (ja) * | 2013-02-21 | 2014-09-08 | Toyota Mapmaster Inc | 表示順設定装置及びその方法、並びに表示順を設定するためのコンピュータプログラム及びコンピュータプログラムを記録した記録媒体 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8494978B2 (en) * | 2007-11-02 | 2013-07-23 | Ebay Inc. | Inferring user preferences from an internet based social interactive construct |
US9978362B2 (en) * | 2014-09-02 | 2018-05-22 | Microsoft Technology Licensing, Llc | Facet recommendations from sentiment-bearing content |
-
2015
- 2015-02-25 WO PCT/JP2015/055481 patent/WO2016135905A1/ja active Application Filing
- 2015-02-25 US US15/551,747 patent/US10671619B2/en active Active
- 2015-02-25 JP JP2017501757A patent/JP6381775B2/ja not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013089109A (ja) * | 2011-10-20 | 2013-05-13 | Nippon Telegr & Teleph Corp <Ntt> | 質問項目推薦装置、質問項目推薦方法及び質問項目推薦プログラム |
WO2013080406A1 (ja) * | 2011-11-28 | 2013-06-06 | Necソフト株式会社 | 対話システム、冗長メッセージ排除方法および冗長メッセージ排除プログラム |
JP2013214294A (ja) * | 2012-03-06 | 2013-10-17 | Okwave:Kk | クライアントシステム及びサーバ |
JP2014164351A (ja) * | 2013-02-21 | 2014-09-08 | Toyota Mapmaster Inc | 表示順設定装置及びその方法、並びに表示順を設定するためのコンピュータプログラム及びコンピュータプログラムを記録した記録媒体 |
Non-Patent Citations (3)
Title |
---|
TOMONORI YOSHIDA ET AL.: "Automated Extraction of Related Term Networks from FAQ Site", IEICE TECHNICAL REPORT, vol. 108, no. 119, 23 June 2008 (2008-06-23), pages 75 - 80 * |
YUTAKA KABUTOYA ET AL.: "Effective Question Recommendation Using Multiple Features for Question Answering Communities", TRANSACTIONS OF INFORMATION PROCESSING SOCIETY OF JAPAN , IPSJ TRANSACTION HEISEI 22 NENDO (2, vol. 3, no. 4, 2 May 2011 (2011-05-02), pages 34 - 47 * |
YUTAKA KABUTOYA ET AL.: "Questions recommendation based on evolution patterns of a QA community", JOURNAL OF THE DBSJ, vol. 8, no. 1, 26 June 2009 (2009-06-26), pages 89 - 94 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6250121B1 (ja) * | 2016-09-16 | 2017-12-20 | ヤフー株式会社 | 地図検索装置、地図検索方法、および地図検索プログラム |
JP2018045636A (ja) * | 2016-09-16 | 2018-03-22 | ヤフー株式会社 | 地図検索装置、地図検索方法、および地図検索プログラム |
EP3327592A1 (en) * | 2016-11-25 | 2018-05-30 | Panasonic Intellectual Property Management Co., Ltd. | Information processing method, information processing apparatus, and non-transitory recording medium |
CN108109616A (zh) * | 2016-11-25 | 2018-06-01 | 松下知识产权经营株式会社 | 信息处理方法、信息处理装置以及程序 |
US11093537B2 (en) | 2016-11-25 | 2021-08-17 | Panasonic Intellectual Property Management Co., Ltd. | Information processing method, information processing apparatus, and non-transitory recording medium |
CN108153800A (zh) * | 2016-12-06 | 2018-06-12 | 松下知识产权经营株式会社 | 信息处理方法、信息处理装置以及程序 |
JP2018092582A (ja) * | 2016-12-06 | 2018-06-14 | パナソニックIpマネジメント株式会社 | 情報処理方法、情報処理装置、及び、プログラム |
CN108153800B (zh) * | 2016-12-06 | 2023-05-23 | 松下知识产权经营株式会社 | 信息处理方法、信息处理装置以及记录介质 |
JP2018173784A (ja) * | 2017-03-31 | 2018-11-08 | 富士通株式会社 | 検索処理方法、検索処理プログラム、および検索処理装置 |
JP2020140692A (ja) * | 2019-03-01 | 2020-09-03 | 楽天株式会社 | 文抽出システム、文抽出方法、及びプログラム |
JP2021039715A (ja) * | 2019-08-30 | 2021-03-11 | バイドゥ オンライン ネットワーク テクノロジー (ベイジン) カンパニー リミテッド | コンテンツ埋込方法、装置、電子デバイス、記憶媒体、及びプログラム |
JP7051190B2 (ja) | 2019-08-30 | 2022-04-11 | バイドゥ オンライン ネットワーク テクノロジー(ペキン) カンパニー リミテッド | コンテンツ埋込方法、装置、電子デバイス、記憶媒体、及びプログラム |
Also Published As
Publication number | Publication date |
---|---|
US20180039633A1 (en) | 2018-02-08 |
JP6381775B2 (ja) | 2018-08-29 |
US10671619B2 (en) | 2020-06-02 |
JPWO2016135905A1 (ja) | 2017-06-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6381775B2 (ja) | 情報処理システム及び情報処理方法 | |
US9836511B2 (en) | Computer-generated sentiment-based knowledge base | |
KR102075833B1 (ko) | 미술 작품 추천 큐레이션 방법 및 시스템 | |
KR102146152B1 (ko) | 관능 평가 방법 및 그 장치 | |
US20200192921A1 (en) | Suggesting text in an electronic document | |
JP6767342B2 (ja) | 検索装置、検索方法および検索プログラム | |
JP4613346B2 (ja) | キーワード抽出方法、キーワード抽出プログラムおよびキーワード抽出装置、並びにメタデータ作成方法、メタデータ作成プログラムおよびメタデータ作成装置 | |
Feuerbach et al. | Enhancing an Interactive Recommendation System with Review-based Information Filtering. | |
JP6928044B2 (ja) | 提供装置、提供方法及び提供プログラム | |
KR20100023630A (ko) | 카테고리 태그 정보를 이용한 웹 페이지 분류 방법, 그 시스템 및 이를 기록한 기록매체 | |
JP6932162B2 (ja) | 地域に基づくアイテム推薦端末装置及びアイテム推薦情報提供方法。 | |
KR101826594B1 (ko) | 지식 구조 기반의 전자책 추천 방법 및 시스템 | |
KR20110038247A (ko) | 키워드 추출 장치 및 방법 | |
US20130332440A1 (en) | Refinements in Document Analysis | |
Al-Saffar et al. | Survey on Implicit Feedbacks Extraction based on Yelp Dataset using Collaborative Filtering | |
KR101265467B1 (ko) | 블로그 문서에서의 경험 문장 추출 방법 및 동사 분류 방법 | |
JP5368900B2 (ja) | 情報提示装置、情報提示方法およびプログラム | |
Balaji et al. | Finding related research papers using semantic and co-citation proximity analysis | |
JP5741242B2 (ja) | プロファイル更新装置およびその制御方法、ならびに、プロファイル更新用プログラム | |
JP7160971B2 (ja) | コンテンツ制御装置 | |
Lerttripinyo et al. | Accommodation recommendation system from user reviews based on feature-based weighted non-negative matrix factorization method | |
JP7037778B2 (ja) | 検索装置および方法 | |
JP7323484B2 (ja) | 情報処理装置、情報処理方法、及びプログラム | |
WO2021095728A1 (ja) | コンテンツ制御装置、コンテンツ制御システム、コンテンツ制御方法、及びコンテンツ制御プログラム | |
JP7008102B2 (ja) | 情報処理装置、情報処理方法、及び情報処理プログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15883201 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2017501757 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15551747 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15883201 Country of ref document: EP Kind code of ref document: A1 |