JP2014085873A - Information processor, information processing system, and question answering method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device, a system, and a method for flexibly answering a fact type question.SOLUTION: A knowledge retrieval answering server 100 includes: an input/output control unit 101 for obtaining a question sentence; a rule type retrieval part 105, a regular expression type retrieval part 106, an entity attribute type retrieval part 107, and a programmed type answer generation part 108 for answering the question sentence; and an answer control part 102 generates an answer result based on the answer content. The input/output control unit 101 outputs an answer result.

Description

  The present invention relates to an information processing apparatus that answers a question, an information processing system, and a question answering method.

  Due to the advancement of user interface using voice recognition technology and touch panel, and the advancement of wireless communication technology, multi-domain question answering systems that answer questions related to various fields from users, not limited to specific fields, are becoming widespread. . The multi-domain question answering system assumes all questions related to the daily life of the user, and answers questions that do not require practical benefits, such as inquiries about data related to entertainment such as movies and general common knowledge such as geographic history, and chats. These pieces of information are acquired from a database management system in the system or a linked external service.

  Since user questions are very diverse, it is common practice to analyze questions and generate answers on the premise that the service (or the partial services that make it up) can answer only certain question types. ing.

  For example, in a car navigation system, the search target is limited to geographical data such as facilities, and the search key specified by the user is limited to only a word or a set of words and the type of target to search (such as “convenience store” or “family restaurant”) doing. In this case, it is not necessary to analyze the input from the user, and the object associated with the type information stored in the system may be searched as it is.

  On the other hand, in the case of a multi-domain type question answering system, it is difficult to limit the question type from the user to a certain form. There are questions such as “Ohson convenience store” and “Looking for a convenience store called Orson” and “Looking for a convenience store called Orson in XX town”. Which question should be set as which search condition for each question is different, and the database preparation method for answering it is also different.

  In addition, there is a technique in which a plurality of answer sentence templates corresponding to a predetermined question are prepared, and answers are extracted based on the plurality of templates (see, for example, Patent Document 1).

JP 2012-79161 A

  However, the method disclosed in Patent Document 1 merely prepares a plurality of answer templates, and is based on the premise that a certain question format is used, and has a problem that it cannot respond to various questions.

  Accordingly, an object of the present invention is to provide an apparatus, a system, and a method that can flexibly answer fact-type questions in order to solve the above-described problems. Here, the fact type question means a question having a specific expression as an answer, unlike a question about a reason or a method.

  In order to solve the above-described problem, an information processing apparatus of the present invention includes a question sentence acquisition unit that acquires a question sentence, an answer unit that includes a plurality of answer units that output all or part of the question sentence, and an answer An answer generating means for generating an answer result based on the output result of the means, and an answer output means for outputting the answer result generated by the answer generating means. The answer means can answer the fact type question of the first pattern. And at least an answering section capable of answering the fact type question of the second pattern.

  The question answering method of the present invention is a question answering method executed by the information processing apparatus, and includes a question sentence obtaining step for obtaining a question sentence, and a plurality of answer units for outputting all or part of the answer of the question sentence. Having an answer step, an answer generation step for generating an answer result based on an output result of the answer step, and an answer output step for outputting the answer result generated by the answer generation step. The answer step is a first pattern. At least an answer section that can answer the fact type question and an answer section that can answer the fact type question of the second pattern.

  According to the present invention, since the answer section for a plurality of fact type questions is provided, it is possible to answer each pattern question. That is, it is possible to flexibly answer various types of fact-type questions.

  In the information processing apparatus of the present invention, the answering unit includes a program answering unit that outputs a program operation result as an answer, and further searches for and retrieves an answer to a definition question that is a question uniquely determined from a question sentence. It includes at least one of a definition type question answering unit that outputs a result or an attribute type question answering unit that searches an answer of an attribute type question that is a question for inquiring about an attribute of a predetermined object, and outputs a search result, When the question sentence is a question corresponding to a definition type question or an attribute type question and includes a word that can be interpreted as requiring a program operation in the question sentence, the answer generation means includes an output result by the program answering unit and a definition type question. You may make it produce | generate the answer based on the output result by any one of an answer part or an attribute type question answer part.

  In this case, with respect to the definition type question and the attribute type question, not only a fixed answer but also an answer requiring a program calculation can be given.

  Further, in the information processing apparatus of the present invention, when the program corresponding to the wording that can be interpreted as requiring the program operation included in the question sentence requires an argument, the program answering unit is provided with the predetermined answer included in the question sentence. Program calculation may be performed using the wording as an argument.

  In this case, since an answer of a program operation using an argument can be made, an answer that can be dynamically changed can be generated using the argument.

  The information processing apparatus of the present invention further includes a question sentence type determining unit that determines whether or not the question sentence is a fact-type question. When the question sentence is not a fact-type question, the question sentence is sent to another device. An answer request may be made.

  In this case, if the question text is not a fact-type question, a request for answering the question text will be sent to another device, so even in the field where it is difficult to prepare data and logic for creating answers in advance. Answers can be generated from existing Web sites. In addition, since it is determined in advance whether or not it is a fact type question, it is possible to prevent the answer process from being performed even though the question is not a fact type question.

  Further, in the information processing apparatus of the present invention, the answer generation means generates an answer result that can distinguish between a case where there is no answer and a case where the answer is an unanswerable question, based on the answer acquired from the answer means.

  In this case, when no answer is obtained, it is possible to determine whether there is no answer or whether the answer is a question that can be handled by the information processing apparatus.

  In the information processing apparatus of the present invention, the answer generation means switches the user interface of the answer result according to the type of the question sentence. In this case, since the interface of the answer result is changed according to the type of the question sentence, the user's understanding can be improved.

  In the information processing apparatus of the present invention, when the supplementary information is included in the result output by the reply means, the reply generation means creates reply information to which the supplementary information is added.

  In this case, for example, additional information such as a title and a unit is added to the object that is being asked in the question sentence, so that consideration for related parties and conversion to a different unit can be performed.

  The information processing system of the present invention is an information processing system including a terminal device and an information processing device, and the information processing device includes a question sentence acquisition unit that acquires a question sentence from the terminal device, and all of the question sentences. Alternatively, an answering unit having a plurality of answering units for outputting some answers, an answer generating unit for generating an answer result based on an output result by the answering unit, and outputting an answer result generated by the answer generating unit to the terminal device Answering means, and the answering means has at least an answering part capable of answering the fact type question of the first pattern and an answering part capable of answering the fact type question of the second pattern. An answer request unit that transmits a question sentence to the information processing apparatus and requests an answer, an answer result acquisition unit that acquires an answer result from the information processing apparatus, and an answer result display unit that displays the answer result , Comprising a.

  According to the present invention, since the answering section for a plurality of patterns of fact type questions is provided, it is possible to answer questions of each pattern. That is, it is possible to flexibly answer various types of fact-type questions.

It is a block diagram which shows the function of an information processing system. It is a hardware block diagram of the knowledge search answer server 100 grade | etc.,. It is an example of an output screen when a normal answer is not obtained. It is an example of an output screen when a normal answer is obtained. 7 is a diagram illustrating an example of data held in a rule-answer table 110. FIG. 7 is a diagram illustrating an example of data held in a regular expression rule table 113. FIG. It is a figure which shows an example of the data currently hold | maintained by the entity-attribute value table. It is a figure which shows an example of the data currently hold | maintained in the attribute display format table. It is a flowchart which shows the process of the question front server 50. FIG. It is a flowchart which shows the process which extracts a reply service candidate. 3 is a flowchart showing processing of a knowledge search answer server 100. It is a flowchart which shows a non-service object determination process. It is a flowchart which shows a search engine utilization suitable determination process. It is a flowchart which shows the reply production | generation process by rule type | mold search. It is a flowchart which shows the reply production | generation process by a regular expression type | mold search. It is a flowchart which shows the reply production | generation process by an entity-attribute type search.

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

  FIG. 1 is a block diagram illustrating functions of the information processing system according to the present embodiment.

  As shown in FIG. 1, the information processing system includes a user terminal 30 such as a mobile phone, a question answer front server 50, a knowledge search answer server 100, a search engine use answer service server 150, and a plurality of other answer servers 60 (other answer servers). 60A, other answer server 60B,. In the present embodiment, the user terminal 30 functions as a terminal device, and the knowledge search answer server 100 functions as an information processing device.

  Here, the user terminal 30, the question answering front server 50, the knowledge search answering server 100, the search engine using answering service server 150, and the other answering server 60 are each connected via a network. A private line from the public communication network and a communication bus in the same physical device may be used, and the physical arrangement thereof does not matter. The functions of the question / answer front server 50 and the knowledge search / answer server 100 do not need to be arranged in a single server device, and the physical configuration is not limited. This includes the case where each is arranged in a separate server device connected via a LAN (Local Area Network) or the Internet.

  FIG. 2 is a hardware configuration diagram of the user terminal 30, the question / answer front server 50, and the knowledge search / answer server 100. As shown in FIG. 2, the user terminal 30, the question answer front server 50, and the knowledge search answer server 100 shown in FIG. 2 are physically a CPU 11, a RAM 12 and a ROM 13, which are main storage devices, and an input device. The computer system includes an input device 14 such as a keyboard and a mouse, an output device 15 such as a display, a communication module 16 that is a data transmission / reception device such as a network card, an auxiliary storage device 17 such as a hard disk or a semiconductor memory, and the like. Each function in FIG. 1 operates the input device 14, the output device 15, and the communication module 16 under the control of the CPU 11 by loading predetermined computer software on the hardware such as the CPU 11 and the RAM 12 shown in FIG. 2. In addition, it 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.

  When the user terminal 30 receives the question text via the input device, the user terminal 30 transmits the question text to the question answer front server 50 via the network and requests the answer to the question text. Then, the user terminal 30 receives the answer result from the question answer front server 50 and displays and outputs the answer result.

  The question / answer front server 50 includes an answer service candidate determination unit 51, an answer generation unit 52, and a service control unit 53.

  The answer service candidate determination unit 51 reads the question sentence received from the user terminal 30 and determines which server of the knowledge search answer server 100 and the plurality of other answer servers 60 should make a question request. In the present embodiment, it is assumed that the answer service candidate determination unit 51 determines the answer request target server based on a word included in the question sentence.

  Specifically, the answer service candidate determination unit 51 sets the knowledge search answer server 100 as an answer candidate server when the ending of the question sentence is “Tell me”. When the ending of the question sentence is “I want to go”, the answer service candidate determination unit 51 sets the other answer server 60A as the answer candidate server. When the question sentence includes “weather”, the answer service candidate determination unit 51 sets the other answer server 60B as the answer candidate server.

  In the present embodiment, the question / answer front server 50 requests the knowledge search / answer server 100 to answer questions about general knowledge such as historical facts in general, and questions about public transportation route searches. The other answer server 60A is requested to answer, and the question related to the weather forecast is requested to the other answer server 60B.

  A plurality of answer candidate servers can be determined. For example, for the question “Tell me the weather in Tokyo”, the knowledge search answer server 100 and the other answer server 60B become answer candidate servers under the above-described determination method.

  The other answer server 60 can take charge of various services other than those described here. For example, a service for searching for music data, a navigation system, or a service for determining an instruction to activate a function such as a mailer installed in the user terminal may be used.

  The service control unit 53 mainly performs data transmission / reception with the user terminal 30 and data transmission / reception with the knowledge search answer server 100. The service control unit 53 receives the answers from the knowledge search answer server 100 and the plurality of other answer servers 60, and determines whether or not to make an answer request to the next answer candidate server according to the answer type. The answer types include three types of cases where a meaningful answer (a normal answer to the question) is obtained, the answer is 0, and the answer is out of the corresponding range.

  If the answer is 0, this indicates that the server that received the question text is within the service range, but there was not enough data to answer it. It is indicated that the question was not within the service range to which the server that received the answer answered.

  Each server that actually prepares an answer corresponding to the question text compares the contents of the question text in detail with the range that each server itself should answer, and makes an answer outside the corresponding range. The question sentence can be transferred without being aware of the detailed correspondence range of each server.

  When an answer indicating that it is out of the corresponding range is acquired from the server that requested the answer, the service control unit 53 transmits an answer request for the question to another answer candidate server, and obtains an answer. For example, after transmitting a question such as “Tell me the weather in Tokyo” to the knowledge search answering server 100 and then obtaining an answer indicating that it is out of the corresponding range, a request for answering the question is sent to the other answering server 60 next. Get an answer.

  As a result of obtaining answers to the question text from all answer candidate servers, the service control unit 53 obtains an answer “out of range” from all servers, and if the answer service candidate determination unit 51 has no answer service candidates. If it is determined, a signal “out of correspondence range” is transferred to the answer generation unit 52. When the answer is 0 or a meaningful answer is obtained, the service control unit 53 transmits the obtained answer to the answer generation unit 52 as it is.

  The answer generation unit 52 creates an answer user interface for transfer to the user terminal 30 in accordance with the answer status obtained from the service control unit 53. The answer user interface may be a GUI (graphical user interface) for outputting to a screen such as a touch panel, read-out sentence data for voice answer, or a combination of the two.

  The answer generation unit 52 displays information (for example, screen information) received from the knowledge search answer server 100 or other answer server 60 on the user terminal 30 as it is when an answer that has some meaning is obtained. . The user terminal 30 presents it to the user in a case where “0 response” and “out of the corresponding range” are distinguished.

  FIG. 3 shows an example in which “0 response” and “out of correspondence range” are displayed on the user terminal 30. FIG. 3A shows an example of a screen when there are no responses. The answer of 0 means that the user has transmitted a question that is a corresponding service but has no answer data. FIG. 3B is an example of a screen when it is out of the corresponding range. If it is out of the corresponding range, it means that the information system cannot answer the requested question. The user terminal 30 may output these character strings by voice. As described above, when no answer is obtained, it can be determined whether there is no answer or whether the answer is a question that can be handled by the knowledge search answer server 100.

  FIG. 4 shows an example when the answer result is normally acquired. FIG. 4A shows an example of answering with characters, and FIG. 4B shows an example of answering using an image. Thus, the user interface may be changed according to the answer result to the question text. Further, the user interface of the answer result may be changed according to the question pattern. Thereby, a user's comprehension degree can be improved.

  Subsequently, returning to FIG. 1, the knowledge search answer server 100 will be described. The knowledge search answer server 100 includes an input / output control unit 101 (question sentence acquisition unit, answer output unit), an answer control unit 102 (answer generation unit), an out-of-service determination unit 103, and a search engine use preference determination unit 104 (question sentence). Type determining means), rule type search unit 105 (definition type question answering unit), regular expression type search unit 106, entity-attribute type search unit 107 (attribute type question answering unit), program type answer generating unit 108 (program answering unit) ), A rule-answer table 110, an entity-attribute value table 111, an attribute display format table 112, and a regular expression rule table 113. The rule type search unit 105, the regular expression type search unit 106, the entity-attribute type search unit 107, and the program type answer generation unit 108 function as an answering unit.

  The input / output control unit 101 transmits / receives data to / from the question / answer front server 50 via the network. Specifically, the answer result acquired from the answer control unit 102 is output, or the question message is transmitted to the answer control unit 102 after receiving the question message transmitted by the user terminal 30. The input / output control unit 101 searches the response result of the question sentence generated by the answer control unit 102 (0 cases, out of the corresponding range, or any of the rule type, regular expression type, entity-attribute type, search engine use answer service) Result) and send the answer result to the question answering front server 50.

  The answer control unit 102 performs control related to the answer generation of the question sentence. Specifically, the non-service target determination unit 103 determines whether or not the question text is out of service, or the search engine usage preference determination unit 104 causes the search engine usage answer service server 150 to answer the question text. Whether the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type search unit 107 are requested to answer questions, the rule type search unit 105, the regular expression An answer result is generated based on the answers acquired from the type search unit 106 and the entity-attribute type search unit 107.

  The answer control unit 102 receives the question text from the input / output control unit 101, and causes the out-of-service determination unit 103 to determine whether or not the question can be answered by the knowledge search answer server 100. When the response control unit 102 acquires the information “out of target range” from the out-of-service determination unit 103, the response control unit 102 transfers the information as a response result to the input / output control unit 101. When the answer control unit 102 does not acquire the information “out of target range”, the answer control unit 102 transmits a question sentence to the search engine use suitability determination unit 104.

  When the response that the answer in the search engine utilization answer service server 150 is suitable is obtained from the search engine suitability determination part 104, the answer control unit 102 transmits a question sentence to the search engine utilization answer service server 150, An answer request is made to the search engine use answer service server 150. The answer control unit 102 transmits the result (0 or more) obtained from the search engine utilization answer service server 150 to the input / output control unit 101 as an answer result.

  When a reply indicating that it is not preferable to use an answer from the search engine use answer service server 150 is obtained, the answer control unit 102 includes a rule type search unit 105, a regular expression type search unit 106, and an entity-attribute type. A question sentence is transmitted to each of the search units 107, and when an answer is obtained from either of them, an answer result is generated based on the obtained answer, and the generated answer result is transmitted to the input / output control unit 101. The answer control unit 102 may obtain answers from a plurality of the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type search unit 107, and generate all the answer results. Alternatively, the answer result may be selectively generated based on a predetermined priority order or the like.

  If a normal answer is not obtained from any of the rule type search unit 105, regular expression type search unit 106, and entity-attribute type search unit 107, the answer control unit 102 sends a response to the search engine utilization answer service server 150. A question sentence is transmitted, and the result (0 or more) is transmitted as an answer result to the input / output control unit 101.

  The non-service target determination unit 103 determines whether the question sentence is a service target. The service out-of-service determination unit 103 receives the question sentence from the answer control unit 102 and determines whether or not the question should be answered by the knowledge search answer server 100 from the answer service candidate determination unit 51 of the question answer front server 50. Judgment.

  Specifically, the out-of-service determination unit 103 includes keywords (such as “cool” and “cute”) that can be understood as not being questions about general knowledge such as historical facts, although “Teach me” is included. Such as those that indicate subjective questions), keywords that indicate chat (such as the name of the user of the user terminal 30), inappropriate topics (such as questions in the adult field or territorial issues) If it is included in the claim, it is determined that the knowledge search answer server 100 is out of service, and that effect is transmitted to the answer control unit 102.

  Based on the contents of the question text, the search engine use preference determination unit 104 determines whether or not the search engine use answer service server 150 is suitable for the answer. The search engine use suitability determination unit 104 receives a question sentence from the answer control unit 102 and should be supported by the knowledge search answer server 100, but is related to a wide range of knowledge that cannot be answered by an answer using a static database or a fixed program. Determine if it is a question. In this embodiment, a question that asks the reason for things is taken as such a question. Specifically, when keywords such as “why”, “why”, “reason”, “reason” are included in the question sentence, it is determined that an answer using a search engine is suitable, and the determination result is sent to the answer control unit 102. Send.

  The search engine utilization answering service server 150 provides a service for crawling a Web site on the Internet, searching from the contents, and answering the result. The search engine utilization answering service server 150 may provide a service using a general search engine provided on the Internet portal page, or provide a user-participation-type Q & A service that is more specialized for use. Anyway. The search engine utilization answer service server 150 is provided for the purpose of answering knowledge that is not intended to prepare an answer by the rule type search function, the regular expression type search function, and the entity-attribute type search function. .

  When the answer control unit 102 transmits the question text to the search engine use answer service server 150, the search result of the website including the contents of the question text is returned from the search engine use answer service server 150. When providing higher accuracy, the search engine utilization answer service server 150 may perform statistical processing such as arranging words that appear in the search result in order of the number of appearances and setting the highest one as the answer.

  The rule type search unit 105 searches for a fixed answer to a question sentence with a specific phrase according to the information accumulated in the rule-answer table 110, and transmits the search result to the answer control unit 102. A question uniquely determined from a question sentence is called a definition question. An example of the rule-answer table 110 is shown in FIG. The rule-answer table 110 includes a title, a unit given to the rule, the answer, and the answer. The rule type search unit 105 searches the rule-answer table 110 using the question sentence as a search condition. Specifically, the rule-type search unit 105 first deletes a specific ending (“Tell me”, “Who”, “What”, etc.) from the question sentence, and then uses the deleted question sentence as a search key. A complete match / forward match search or partial match search with the rules of the answer table 110 is performed.

  When there is a corresponding record, the rule type search unit 105 extracts an answer included in the record and a title / unit of the answer, and transmits the extracted information to the answer control unit 102. The response control unit 102 generates a response result based on the information acquired from the rule type search unit 105. The answer control unit 102, for example, examines [question text] and the answer is [answer] [answer title / unit]. Is generated as an answer result.

  As a specific example, when the rule type search unit 105 receives a question sentence “What is the highest mountain in the world?”, The rule type search unit 105 first deletes the ending of the question sentence and says “the highest mountain in the world”. Create a search key. When searching for the rule in the rule-answer table 110 using the search key, the rule-type search unit 105 searches for a record whose answer is “Everest” and which has no title or unit for the answer. The rule type search unit 105 transmits an answer based on the searched record to the answer control unit 102. The answer result generated by the answer control unit 102 based on the content transmitted by the rule type search unit 105 is “What is the highest mountain in the world, and the answer is Everest?”.

  Some of the records in the rule-answer table 110 have a program to be activated for an answer and an argument to be delivered to the answer. This means that it is not fixed data, and it is necessary to generate an answer using a specific program by the program type answer generation unit 108 described later. As a specific example, when the rule type search unit 105 accepts a question sentence “what time is it”, the rule type search unit 105 causes the program type answer generation unit 108 to start the program 1 with an argument “Japan”. Request an answer to get. As described above, the rule-type search unit 105 includes a word that can be interpreted as requiring a program operation when the answer of the rule in the rule-answer table 110 corresponding to the question requires a program answer. If so, an answer is requested by the program answer generation unit 108. In the example of FIG. 5, the argument is defined in the rule-answer table 110, but the wording corresponding to the argument may be extracted from the question sentence. An example of processing for determining the argument part from the question sentence will be described later.

  When the rule type search unit 105 searches for a record having an activation program or an argument as described above, the rule type search unit 105 transmits the program and the argument included in the record to the program type answer generation unit 108, and the program type answer generation unit The information acquired from 108 is edited into a response form and transmitted to the response control unit 102.

  The regular expression type search unit 106 makes an answer to a question sentence described in a so-called regular expression type. Specifically, the regular expression type search unit 106 generates an answer to the question sentence obtained from the answer control unit 102 with reference to information stored in the regular expression rule table 113, and the answer control unit To 102. Here, the regular expression rule table 113 is shown in FIG. As illustrated in FIG. 6, the regular expression rule table 113 stores “regular expression pattern”, “answer generation rule”, and “title / unit for answer” in association with each other. The regular expression type search unit 106 uses a processing engine based on a notation method that realizes a flexible pattern match called a regular expression, and activates a program corresponding to the pattern when it matches the pattern. The regular expression processing engine is provided as a standard library in various general-purpose programming languages such as Java (registered trademark), and it is assumed that the regular expression processing unit 106 in the present embodiment also uses them as they are.

  The regular expression type search unit 106 first determines whether there is a regular expression pattern in the regular expression rule table 113 corresponding to the question sentence. The regular expression pattern on the first line in FIG. 6 means that a match is made when an arbitrary character string is followed by a character string “What time is it now?”. For example, if the question “What time is London now?” Is entered, this pattern is matched.

  In addition, the part of the “regular expression pattern” enclosed in parentheses means that it is used as an argument to the program corresponding to the regular expression pattern. In this example, “London” is used as an argument of the answer generation rule.

  When there is a regular expression pattern in the regular expression rule table 113 corresponding to the question sentence, the regular expression type search unit 106 instructs the program type answer generation unit 108 to start a program corresponding to the regular expression pattern. Notice.

  The program-type answer generation unit 108 activates a program specified with an argument and outputs an answer. In the above example, the time of the corresponding city is generated (such as “9:59”) and notified to the regular expression search unit 106.

  When the regular expression type search unit 106 obtains an answer from the program type answer generation unit 108, the regular expression type search unit 106 transmits the answer to the answer control unit 102. The answer control unit 102 generates an answer result for output based on the obtained answer. The answer control unit 102, for example, similarly to the rule type search unit 105, “I investigated [question text]. The answer is [the answer obtained from the program type answer generation unit] [the title / unit of the answer].” Are created using the question text / answer / title of the answer / unit.

  The entity-attribute type search unit 107 answers to a question sentence (attribute type question) for inquiring about an attribute of a predetermined object. Specifically, the entity-attribute type search unit 107 uses the key information (entity search key) representing the target object that the user wants to inquire from the question text as well as the key information (side of the target object) from the question text ( Attribute search key) is extracted.

  Various methods can be considered for these extractions. In this embodiment, in the case of a combination of keywords, the entity-attribute type search unit 107 searches the first keyword for the entity search key and the second keyword for the attribute search. Judge as a key. For example, in the case of the keyword “◯ △ director”, “○ △” is determined as an entity search key, and “director” is determined as an attribute search key.

  In addition, when natural text questions are targeted, attributes that can be accepted by the system in a form that associates attributes with rules of the form "who is the director of XX" (XX is an arbitrary character string) There may be a method that prepares and uses the ○○ portion as an entity search key only when the pattern matches the pattern, and uses the attribute associated with the matched pattern as the attribute search key of the question.

  After extracting these search keys, the entity-attribute type search unit 107 searches the value corresponding to the answer with reference to the entity-attribute value table 111. An example of the entity-attribute value table 111 is shown in FIG. As illustrated in FIG. 7, the entity-attribute value table 111 manages information in which “target”, “attribute”, and “value” are associated with each other.

  The entity-attribute type search unit 107 extracts rows using an entity search key as a partial match search condition for an entity and an attribute search key as a complete search condition for an attribute. In the case of the question in which the entity search key is “◯ △” and the attribute search key is “director”, AAAAAA, “◯ △ and △ ×” are assigned as the director of “◯ △ and the secret room” in the example of FIG. 7. CCBBCC is extracted as the director of BBBBBBB as the director of "The Prisoner of Nothing" and the director of "O △ and the Deathly Hallows".

  Subsequently, the entity-attribute type search unit 107 extracts records from the attribute display format table 112 using the attributes searched from the entity-attribute value table 111 as keys. An example of the attribute display format table 112 is shown in FIG. The attribute display format table 112 manages information in which “attribute”, “title / unit for entity”, “title / unit for value”, and “program answer presence / absence” are associated with each other. The entity-attribute type search unit 107 transmits the record information extracted from the entity-attribute value table 111 and the attribute display format table 112 to the answer control unit 102 as an answer.

  The answer control unit 102 generates a sentence in a form such as “[attribute] of [extracted entity] [entity title / unit] is [value] [value title / unit]”. In the above example, “AAAAA is the director of the secret room with ○ △.” “The director of the prisoner of ○ △ and △ × is BBBBBBB.” “The director of the secret treasure with ○ △ is CCCCCC. "San." Is the answer.

  Further, when a specific attribute search key is extracted from the question sentence, the entity-attribute type search unit 107 notifies the program type answer generation unit 108 to that effect, and the program type answer generation unit 108 answers the program. You may get For example, when the question sentence is “What is the next full moon?”, The entity-attribute type search unit 107 specifies a record including “full moon” as an attribute in the attribute display format table 112 of FIG. Since the program name is included in the presence / absence of the program response of the record “”, the response using the program 3 is requested to the program type response generation unit 108 with the argument “next”. When the question sentence is “What is Thursday next week?”, The entity-attribute type search unit 107 includes “next week” that is information about the week and “Thursday” that is the day of the week in the question sentence. 8 determines that a date is requested from the day of the week, identifies a record whose attribute display format table 112 in FIG. 8 has the attribute “day of the week to date”, and the program name of the record includes the program name. Therefore, the program type response generation unit 108 is requested to make an answer using the program 4 with the arguments as “next week” and “Thursday”. When the question sentence is “What day is next Thursday?”, The entity-attribute type search unit 107 determines that a request for a holiday / anniversary is requested from the word “what day”, A record whose attribute in the attribute display format table 112 of FIG. 8 is “Holiday / Anniversary” is specified, and the program name is included in the presence / absence of the program response of the record. Then, an argument using “next Thursday” is requested and an answer using the program 5 is requested.

  As described above, when the entity-attribute type search unit 107 can determine that the program answer is necessary based on the information on the presence / absence of the program answer in the record of the attribute display format table 112 corresponding to the question sentence, that is, the program calculation is performed on the question sentence. Is included, a response by the program type answer generation unit 108 is requested.

  Note that the processing for determining the argument portion may use a regular expression pattern in the regular expression rule table 113. For example, when a pattern of “XX is what day” is matched, a rule that exists second in the regular expression rule shown in FIG. 6 is activated, and subsequent argument processing is performed.

  The program type answer generation unit 108 is given a program identifier and an argument from each rule of the rule type search, the regular expression type search, and the entity-attribute type search, starts the corresponding program, and calls the answer obtained by calculation. Return to original. The program-type answer generation unit 108 dynamically generates an answer to a question that cannot be answered with fixed data in advance and returns it to the caller.

  For example, a program that accepts a city name as an argument and returns the current time of the city, or a program that accepts an absolute or relative date as an argument and returns a holiday or anniversary on that date can be considered. Note that the program-type answer generation unit 108 may simply output the program operation result without acquiring an argument. For example, a program that outputs the current time without an argument is applicable.

  It should be noted that the service out-of-service determination unit 103, the search engine use preference determination unit 104, the rule type search unit 105, the regular expression search unit 106, and the entity-attribute search unit 107 respectively perform non-service determination and search engine use. Although it has been described that the preference determination, the rule type search, the regular expression type search, and the entity-attribute search are forwarded in this order, a part of them may be collectively implemented as one function. . For example, a flag indicating that the service is not applicable or a flag indicating that the search engine is suitable for use is set in a part of the rules of the rule type search, and if it is matched, A question sentence may be transmitted.

  In addition, in order for the entity-attribute type search 107 to perform an entity-attribute type search, an attribute indicating non-service target and an attribute indicating that it is preferable to use a search engine utilization answer service are displayed in the entity-attribute value table 111. Alternatively, it may be held in the attribute display format table 112 or the like.

  Next, operations of the question / answer front server 50 and the knowledge search / answer server 100 configured as described above will be described.

  First, the operation of the question / answer front server 50 will be described with reference to the flowchart shown in FIG.

  When receiving the question message from the user terminal 30, the service control unit 53 of the question answer front server 50 receives the question message (step S1).

  Then, the answer service candidate determination unit 51 determines an answer service candidate (step S2). Details of the answer service candidate determination process will be described later.

  Subsequently, the service control unit 53 transmits a question sentence to the server selected as a candidate by the answer service candidate determination unit 51 and makes an answer request (step S3).

  When the service control unit 53 obtains the answer result from the server that has made the answer request, if the answer is obtained or the answer is 0 (step S4; Yes), the answer generating unit 52 sends the answer result to the user terminal. The user terminal 30 displays and outputs the answer result (step S6).

  If the service control unit 53 obtains the answer result from the server that made the answer request, no answer is obtained (step S4; No), and if there are other answer service candidates (step S5; Yes), step When the process moves to S3 and there are no other answer service candidates (step S5; No), the answer generation unit 52 transmits information indicating that the answer cannot be made to the user terminal 30 (step S7).

  Next, the process of determining the answer service candidate by the answer service candidate determining unit 51 will be described with reference to the flowchart shown in FIG.

  The answer service candidate determination unit 51 first receives a question sentence (step S21). Next, when the question sentence includes “Tell me” (step S22; Yes), the answer service candidate determination unit 51 adds the knowledge search answer server 100 to the answer service candidate (step S23), and adds it to the question sentence. When “I want to go” is included (step S24; Yes), the other answer server A60A is added to the answer service candidate (step S25), and when “weather” is included in the question sentence (step S26; Yes) ), The other answer server B60B is added to the answer service candidates (step S27). If none of the conditions in step S22, step S24, and step S26 is satisfied, the answer service candidate determination unit 51 notifies the service control unit 53 that there is no answer service candidate.

  Next, the processing of the knowledge search answer server 100 will be described using the flowcharts shown in FIGS.

  First, the overall processing of the knowledge search answer server 100 will be described using the flowchart shown in FIG.

  First, the input / output control unit 101 accepts an answer request from the question answer front server 50 and accepts a question sentence (step S31).

  When the input / output control unit 101 receives the question text, the answer control unit 102 performs control related to the answer. The answer control unit 102 causes the non-service target determination unit 103 to determine whether or not the content of the question text is not a service target (step S32). Details of the non-service determination will be described later.

  When the non-service determination unit 103 determines that the knowledge search answer server 100 is out of service (step S33; Yes), the response control unit 102 generates information indicating that the service is out of target as a response result, and inputs and outputs The control unit 101 outputs the answer result to the question answer front server 50 as an answer result (step S41).

  When the non-service target determination unit 103 determines that the knowledge search answer server 100 is a service target (step S33; No), the search engine use preference determination unit 104 determines whether the search engine use is preferable ( Step S35). Details of the process for determining whether or not the use of the search engine is suitable will be described later.

  When the search engine use preference determination unit 104 determines that the search engine use is suitable (step S36; Yes), the answer control unit 102 requests the search engine use answer service server 150 to answer a question (step step). S37). The answer control unit 102 obtains an answer from the search engine use answer service server 150 and generates the obtained answer as an answer result. The input / output control unit 101 transmits the answer result to the question front server 50 (step S41).

  When the search engine usage preference determination unit 104 determines that the search engine usage is not suitable (step S36; No), the answer control unit 102 includes the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type. Each search unit 107 is caused to search for an answer (step S38). As described above, the knowledge search answer server 100 has a plurality of types of search units, and thus can correspond to a plurality of question patterns.

  Details of the search processing of the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type search unit 107 will be described later.

  When the answer control unit 102 causes each of the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type search unit 107 to perform an answer search, one or more answers are obtained (step S39; Yes) ), The answer control unit 102 generates an answer result from the answer, and the input / output control unit 101 transmits the answer result to the question answer front server 50 (step S41).

  When the answer control unit 102 causes the rule type search unit 105, the regular expression type search unit 106, and the entity-attribute type search unit 107 to perform an answer search, no answer is obtained (step S39; No). The engine utilization answer service server 150 is requested to answer (step S40). The answer control unit 102 generates an answer result based on the answer acquired from the search engine use answer service server 150. For example, when a response indicating that there are no answers is obtained from the search engine utilization response service server 150, the response control unit 102 generates a response result indicating that information. The input / output control unit 101 transmits the answer result generated by the answer control unit 102 to the question answer front server 50 (step S41).

  Next, the non-service target determination process will be described with reference to the flowchart shown in FIG.

  The out-of-service determining unit 103 first acquires a question sentence from the answer control unit 102 (step S51). Next, the service out-of-service determining unit 103 determines that a question keyword includes a subjective keyword (step S52; Yes), and a question sentence includes a keyword indicating chat (step S53; Yes). If the question sentence contains an inappropriate keyword (step S54; Yes), the answer control unit 102 is notified that the question sentence is out of service (step S55).

  Then, the non-service determination unit 103 is a case where a subjective keyword is not included in the question sentence (Step S52; No), and a case where a keyword indicating chat is not included in the question sentence (Step S53). ; No), and when an inappropriate keyword is not included in the question text (step S54; No), the answer control unit 102 is notified that it is a service target (step S56).

  Next, processing for determining whether or not the search engine use preference unit 104 is suitable for use of the search engine will be described with reference to the flowchart shown in FIG.

  First, the search engine use preference determination unit 104 acquires a question sentence from the answer control unit 102 (step S61). Next, when the question sentence includes a keyword indicating the reason (step S62; No), the search engine use preference determination unit 104 notifies the answer control unit 102 that the search engine use is not suitable (step S63). If the keyword indicating the reason is included in the question text (step S62; Yes), it is notified that the answer by the search engine utilization answer service is suitable (step S64). As described above, the search engine usage preference determination unit 104 determines whether or not the search engine usage is suitable, so that it is possible to prevent the answer processing from being performed even though the question is not a fact type question. be able to.

  Next, rule type search processing performed by the rule type search unit 105 will be described with reference to the flowchart shown in FIG.

  First, the rule type search unit 105 receives a question sentence from the answer control unit 102 (step S71), and then the rule type search unit 105 deletes a specific ending (step S72) and searches for a predetermined part of the question sentence. The rule-answer table 110 is searched as a key (step S73).

  The rule type search unit 105 terminates the processing when the answer cannot be searched (step S74; No), and when the answer can be searched (step S74; Yes), retrieves the answer and the title / unit of the answer ( Step S75).

  When the question sentence does not include information that requires a program answer (step S76; No), the rule type search unit 105 transmits the information extracted in step S75 to the answer control unit 102, and the answer control unit 102 Then, an answer result is generated based on the acquired information (step S78).

  When the question sentence includes information that requires a program answer (step S76; Yes), the rule type search unit 105 acquires information from the program type answer generation unit 108 (step S77), and the rule type search unit 105 transmits the acquired information and the information extracted in step S75 to the response control unit 102, and the response control unit 102 generates a response result based on the acquired information (step S78). In this manner, the knowledge search answer server 100 can handle not only fixed answers but also answers that require program operations for definition questions.

  Next, regular expression type search processing performed by the regular expression type search unit 106 will be described with reference to the flowchart shown in FIG.

  First, the regular expression type search unit 106 obtains a question sentence from the answer control unit 102 (step S81), and then the regular expression type search unit 106 refers to the regular expression table rule table 113 to determine the rules of the regular expression table. Is applied to the question sentence (step S82).

  If the rule in the regular expression rule table 113 is not included in the question sentence (step S83; No), the regular expression search unit 106 ends the process, and the rule in the regular expression rule table 113 is included in the question sentence. If yes (step S83; Yes), the identifier of the answer program and the title / unit of the answer are extracted (step S84), information is obtained from the program-type answer generator 108 (step S85), and the obtained information is answered. Transmit to the control unit 102. Then, the answer control unit 102 generates an answer result (step S86).

  Next, entity attribute type search processing performed by the entity-attribute type search unit 107 will be described with reference to the flowchart shown in FIG.

  First, the entity-attribute type search unit 107 receives a question sentence from the answer control unit 102 (step S91). Then, the entity-attribute type search unit 107 extracts an entity search key and an attribute search key from the question sentence (step S92), and searches for entity attribute information (step S93).

  If the entity-attribute type search unit 107 cannot search for an answer (step S94; No), the process ends, and if the answer can be searched (step S94; Yes), the entity title / unit, and the value title -A unit is taken out (step S95).

  The entity-attribute type search unit 107 transmits the information extracted in step S95 to the answer control unit 102 when the question text does not include information that requires a program answer (step S96; No). 102 generates an answer result based on the acquired information (step S98).

  The entity-attribute type search unit 107 acquires information from the program type answer generation unit 108 (step S97) when the question sentence includes information that requires a program answer (step S96; Yes). The attribute type search unit 107 transmits the acquired information and the information extracted in step S95 to the response control unit 102, and the response control unit 102 generates a response result based on the acquired information (step S98). As described above, the knowledge search answer server 100 can handle not only fixed answers but also answers that require program operations for attribute type questions.

  Next, the operation and effect of the knowledge search answer server 100 according to the present embodiment will be described. According to the knowledge search answer server 100, the input / output control unit 101 acquires a question sentence, and the rule type search unit 105, the regular expression type search unit 106, the entity-attribute type search unit 107, and the program type answer generation unit 108 The answer of the question is answered, the answer control unit 102 generates an answer result based on the answer content, and the input / output control unit 101 outputs the answer result.

  As described above, the knowledge search answering server 100 has answer parts for a plurality of patterns of fact-type questions (for example, definition-type questions and attribute-type questions), and therefore can answer questions of each pattern. . That is, it is possible to flexibly answer various types of fact-type questions.

  When the question sentence is a definition type question or an attribute type question, and further includes a word that can be interpreted as requiring a program operation in the question sentence, the answer control unit 102 receives the answer from the program type answer generation unit 108 and the rule type search unit 105. Alternatively, an answer result based on the answer by any one of the entity-attribute type search unit 107 is generated.

  In this case, with respect to the definition type question and the attribute type question, not only a fixed answer but also an answer requiring a program calculation can be handled.

  The program-type answer generation unit 108 outputs a program operation result corresponding to the argument. In this case, since an answer of a program operation using an argument can be made, an answer that can be dynamically changed can be generated using the argument.

  The search engine use preference determination unit 104 determines whether the question sentence is a fact type question. When the search engine use preference determination unit 104 determines that the question text is not a fact type question, the knowledge search answer server 100 requests the search engine use answer service server 150 to answer the question text. In this case, since the knowledge search answer server 100 makes a request for answering the question sentence to another device when the question sentence is not a fact type question, it is difficult to prepare data and logic for preparing the answer in advance. In addition, an answer can be generated from a Web site existing on the Internet. In addition, since it is determined in advance whether or not it is a fact type question, it is possible to prevent the answer process from being performed even though the question is not a fact type question.

  The answer control unit 102 generates an answer result that can identify when there is no answer and when the answer is an unanswerable question. In this case, when an answer is not obtained, the user can determine whether there is no answer or whether the answer is a question that can be handled by the knowledge search answer server 100.

  The answer control unit 102 switches the user interface of answer results according to the type of question sentence. In this case, since the interface of the answer result is changed according to the type of the question sentence, the user's understanding can be improved.

  When the answer control unit 102 acquires additional information such as a value or a title from the rule type search unit 105 or the entity-attribute type search unit 107, the response control unit 102 generates a response result to which the additional information is also added. In this case, since the title or unit is added as additional information for the object being heard in the question sentence, consideration for the parties concerned or conversion to a different unit can be performed.

(Other embodiments)
In the above embodiment, the case where the question answer front server 50 and the knowledge search answer server 100 are configured separately has been described. However, the question answer front server 50 and the knowledge answer server 100 may be integrated.

DESCRIPTION OF SYMBOLS 30 ... User terminal, 50 ... Question answer front server, 51 ... Answer service candidate determination part, 52 ... Answer generation part, 53 ... Service control part, 60 ... Other answer server, 100 ... Knowledge search answer server, 101 ... Input / output control , 102 ... Reply control part, 103 ... Out-of-service determination part, 104 ... Search engine use suitability determination part, 105 ... Rule type search part, 106 ... Regular expression type search part, 107 ... Entity-attribute type search part, 108 ... Program type answer generation unit, 110... Rule-answer table, 111.

In order to solve the above-described problem, an information processing apparatus of the present invention includes a question sentence acquisition unit that acquires a question sentence, an answer unit that includes a plurality of answer units that output all or part of the question sentence, and an answer An answer generating means for generating an answer result based on the output result of the means, and an answer output means for outputting the answer result generated by the answer generating means. The answer means can answer the fact type question of the first pattern. and Do reply section, and a reply possible answers portion facts type questions of the second pattern at least chromatic, the answer means include a program reply unit for outputting the program operation result as an answer, further uniquely from the question sentence Search for answers to definition questions that are defined questions, search for answers to definition questions that output the search results, or search for answers to attribute questions that ask questions about the attributes of a given object. It includes at least one of attribute type question answering sections that output results, and the question sentence is a question corresponding to a definition type question or an attribute type question, and includes a word that can be interpreted as requiring a program operation in the question sentence. In this case, the answer generation means generates an answer based on the output result by the program answering unit and the output result by any one of the question definition type question answering unit or the attribute type question answering unit .

The question answering method of the present invention is a question answering method executed by the information processing apparatus, and includes a question sentence obtaining step for obtaining a question sentence, and a plurality of answer steps for outputting all or part of the answer of the question sentence. Having an answer step, an answer generation step for generating an answer result based on an output result of the answer step, and an answer output step for outputting the answer result generated by the answer generation step. The answer step is a first pattern. a reply unit which can answer the facts type questions, see at least contains the answer possible answers portion facts type questions second pattern, the answer step includes a program answer step of outputting a program operation result as a reply, further , A definition type question answering step that searches for answers to a definition type question that is a question uniquely determined from a question sentence, and outputs a search result It includes at least one of attribute type question answering steps for searching for an answer to an attribute type question that is a question for inquiring an attribute of a predetermined object, and outputting a search result, and the question sentence is a definition type question or an attribute type question. If the question sentence contains a word that can be interpreted as requiring a program operation, the answer generation step will output either the program answer step, the question definition type question answer step, or the attribute type question answer step. An answer based on the output result is generated .

The information processing system of the present invention is an information processing system including a terminal device and an information processing device, and the information processing device includes a question sentence acquisition unit that acquires a question sentence from the terminal device, and all of the question sentences. Or an answering means having a plurality of answering parts for outputting some answers, an answer generating means for generating an answer result based on an output result by the answering means, and an answer for outputting the answer result generated by the answer generating means to the terminal device Output means, and the answer means has at least an answer section that can answer the fact type question of the first pattern and an answer section that can answer the fact type question of the second pattern . It includes a program answering section that outputs the results of program calculations as answers, and further searches for answers to definition questions that are questions that are uniquely determined from the question text, and outputs the search results. Or at least one of attribute type question answering units that output a search result, and the question sentence is a definition type question or If the question is a question corresponding to an attribute-type question and the question sentence includes a word that can be interpreted as requiring a program operation, the answer generation means outputs the result output by the program answer part and the question definition-type question answer part or the attribute type question answer The terminal device generates a response based on the output result of any one of the units , and the terminal device transmits a question sentence to the information processing device, and obtains a response result from the information processing device and a response requesting unit that makes a response request Answer result acquisition means for displaying the answer results and answer result display means for displaying the answer results.

Claims (9)

A question sentence acquisition means for acquiring a question sentence;
An answering means having a plurality of answering sections for outputting all or part of the answer of the question text;
Answer generating means for generating an answer result based on the output result by the answer means;
An answer output means for outputting the answer result generated by the answer generating means,
The information processing apparatus according to claim 1, wherein the answering unit includes at least an answering unit capable of answering a fact type question of a first pattern and an answering unit capable of answering a fact type question of a second pattern. The answering means includes a program answering unit that outputs a program operation result as the answer, and further searches for answers to a definition question that is a question uniquely determined from a question sentence, and outputs a search result Or at least any one of an attribute type question answering unit that retrieves an answer to an attribute type question that is a question for inquiring about an attribute of a predetermined object, and outputs a search result,
When the question sentence is a question corresponding to a definition type question or an attribute type question and includes a word that can be interpreted as requiring a program operation in the question sentence, the answer generating means includes an output result by the program answering unit, The information processing apparatus according to claim 1, wherein an answer based on an output result of any one of the question definition type question answering unit and the attribute type question answering unit is generated. When the program corresponding to the wording that can be interpreted as requiring the program operation included in the question sentence requires an argument,
The information processing apparatus according to claim 2, wherein the program answering unit performs a program operation using a predetermined word included in the question sentence as an argument. A question sentence type determining means for determining whether or not the question sentence is a fact type question,
The information processing apparatus according to any one of claims 1 to 3, wherein when the question sentence is not a fact type question, a request for answering the question sentence is sent to another apparatus.   5. The answer generation unit generates an answer result that can distinguish between a case where there is no answer and a case where the answer cannot be answered based on the answer acquired from the answer unit. The information processing apparatus according to any one of the above.   The information processing apparatus according to any one of claims 1 to 5, wherein the answer generation unit switches a user interface of answer results according to a question sentence type.   The answer generation unit generates the answer information to which the additional information is added when the additional information is included in the result output by the answering unit. The information processing apparatus described. An information processing system comprising a terminal device and an information processing device,
The information processing apparatus includes:
Question sentence acquisition means for acquiring a question sentence from the terminal device;
An answering means having a plurality of answering sections for outputting all or part of the answer of the question text;
Answer generating means for generating an answer result based on the output result by the answer means;
Answer output means for outputting the answer result generated by the answer generating means to the terminal device,
The answering means includes at least an answering unit capable of answering a fact type question of the first pattern and an answering unit capable of answering a fact type question of the second pattern,
The terminal device
An answer request means for sending a question sentence to the information processing apparatus and making an answer request;
An answer result obtaining means for obtaining an answer result from the information processing apparatus;
An answer result display means for displaying the answer result;
An information processing system comprising: A question answering method executed by an information processing apparatus,
A question sentence acquisition step for acquiring a question sentence;
An answer step having a plurality of answer sections for outputting all or part of the answer of the question sentence;
An answer generation step of generating an answer result based on the output result of the answer step;
An answer output step of outputting the answer result generated by the answer generating step,
The answering step includes at least an answering part capable of answering a fact type question of the first pattern and an answering part capable of answering a fact type question of the second pattern.

Images