Patent Application
20170249314
This application is based upon and claims the benefit of priority of the prior Japanese Patent Application No. 2016-035461, filed on Feb. 26, 2016, the entire contents of which are incorporated herein by reference.
The embodiments discussed herein are related to apparatus and method to determine a predicted reliability of searching for an answer to question information.
Providers who provide service to users (hereinafter also simply referred to as providers) build and operate business systems (hereinafter, also referred to as, information processing systems) suitable for usage purposes in order to provide various kinds of services to the users, for example. When an information processing system receives a question text (hereinafter, also referred to as question information) on the service from a user, for example, the information processing system searches a storage unit in which answer texts to question texts (hereinafter, also referred to as answer information) are stored to find an answer text to the received question text. The information processing system then transmits the searched-out answer text to the user,
When searching for an answer text as described above, the information processing system segments the received question text into morphs to generate a keyword group including multiple keywords, for example. The information processing system then extracts an answer text that includes a large number of keywords among the keywords in the generated keyword group, from the multiple answer texts stored in the storage unit, for example. This enables the provider to transmit to the user the answer text to the question text received from the user (for example, refer to Japanese Laid-open Patent Publication Nos. 2002-334107, 09-81578, 2003-91556, and 2007-102723).
According to an aspect of the invention, an apparatus stores teacher data and supplementary information, where the teacher data includes first question information and first answer information, each piece of the first question information indicates a question about a predetermined subject, each piece of the first answer information is associated with a piece of the first question information and indicates an answer that is responsive to the piece of the first question information, and each piece of the supplementary information is associated with one or more keywords that are used within the first question information or the first answer information in connection with the each piece of supplementary information. The apparatus extracts first keywords from the teacher data, and adjusts a calculation parameter including parameter-values that are each associated with one of pieces of the supplementary information and used for calculating a predicted-reliability, based on the supplementary information associates with the first keywords, and right/wrong information indicating whether each piece of the first answer information is a right answer to a piece of the first question information associated with the each piece of the fast answer information, where the predicted-reliability indicates a likelihood that each piece of the first answer information is an answer that is responsive to a piece of the first question information associated with the each piece of the first answer information. When outputting plural pieces of second answer information in response to new question information, the apparatus calculates the predicted-reliability of each piece of the second answer information, based on the adjusted calculation parameter, by using the supplementary information associated with second keywords that are extracted from the new question information and the each piece of the second answer information.
The object and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the claims.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention, as claimed.
When searching out an answer text to the question text received from the user, the information processing system as described above outputs the searched-out answer text to an output device viewable by the user, for example. Then, when searching out multiple answer texts, the information processing system preferentially outputs the answer texts determined to be more appropriate, for example.
However, the answer text that the user seeks for does not match the answer text that the information processing system determines to be more appropriate for the question text, in some cases. Moreover, the user may read only the most-preferentially outputted answer text (for example, the answer text outputted at a position most-easily viewed by the user in the output device) among the answer texts outputted to the output device, in some cases. Accordingly, the information processing system may fail to allow the user to read the answer text that the user seeks for, in some cases.
It is preferable to appropriately determine the priority for outputting searched-out results.
[Configuration of Management Device and Physical Machine]
When the information processing device 1 receives question information transmitted from the provider terminal 11 that is a terminal used by a provider, the information processing device 1 searches for answer information to the received question information (answer information that includes information for solving a question included in the received question information). The information processing device 1 then transmits the searched-out answer information to the provider terminal 11.
The provider terminals 11 are terminals used by the providers, and each transmit question information to the information processing device 1, for example. Specifically, for example, the provider terminal 11 extracts a part of the content described in an e-mail (for example, e-mail in which a content of inquiry related to a service is described) that is transmitted from a user, and transmits the extracted part of the content as question information to the information processing device 1. Moreover, the provider terminal 11 transmits a content (for example, inquiry content related to a service) inputted by a person in charge who was contacted by phone from a user as question information, to the information processing device 1, for example.
[Search for Answer Information]
Next, a search for answer information will be described.
As illustrated in
When the information processing device 1 receives the question information transmitted by the provider terminal 11, the information processing device 1 then searches for answer information to the received question information ((2) of
Thereafter, the information processing device 1 transmits the searched-out answer information to the provider terminal 11 ((3) of
Here, when multiple searched-out pieces of answer information are present, for example, the information processing device 1 causes the provider terminal 11 to more preferentially output answer information determined to be more appropriate by the information processing device 1. However, as illustrated in
To address this, the information processing device 1 in this embodiment extracts keywords from question information (hereinafter, also referred to as first question information) and answer information (hereinafter, also referred to as first answer information), which are included in teacher data. The information processing device 1 then executes machine learning on a calculation parameter including parameter-values used for calculating a predicted reliability of the first answer information, which indicates how much the first answer information is likely to be an answer that is responsive to the first question information. For example, the information processing device 1 executes machine learning, based on supplementary information associated with keywords extracted from first question information, supplementary information associated with keywords extracted from the first answer information, and right/wrong information indicating whether the first answer information is a right answer to the first question information. Further, the supplementary information is information identifying a group of keywords having meanings falling under the same concept, in other words, is information on a higher-level concept of the keywords.
Thereafter, when the information processing device 1 outputs multiple pieces of answer information (hereinafter, also referred to as second answer information) to inputted new question information (hereinafter, also referred to as second question information), the information processing device 1 calculates the predicted-reliability of each of the multiple pieces of second answer information. Specifically, the information processing device 1 calculates the predicted-reliability of each of the multiple pieces of the second answer information with a calculation parameter obtained through the machine learning, based on supplementary information associated with keywords extracted from the second question information, and supplementary information associated with keywords extracted from the piece of the second answer information.
In other words, for example, the provider selects in advance question information to be highly likely received from the provider terminal 11 as first question information. Moreover, when a search with'the selected first question information is performed, the provider selects answer information to be desirably searched out as first answer information. In addition, for example, the provider creates teacher data in which the selected first question information and the selected first answer information, and right/wrong information indicating that the selected first answer information is an appropriate answer (right answer) to the first question information are associated with each other. Moreover, when a search with the selected first question information is performed, the provider selects answer information to be desirably searched out as different first answer information. Further, for example, the provider creates teacher data in which the selected first question information and the different first answer information, and right/wrong information indicating that the selected different first answer information is not an appropriate answer (wrong answer) to the first question information are associated with each other. Thereafter, the information processing device 1 executes machine learning by associating the first question information, the first answer information, and the right/wrong information, which are included in the teacher data with each other.
This allows the information processing device 1 to execute machine learning for first question information while distinguishing first answer information that the user seeks for from first answer information (different first answer information) that the user does not seek for.
Meanwhile, when the information processing device 1 outputs multiple pieces of second answer information that are results of the search with the inputted second question information, the information processing device 1 refers to a calculation parameter obtained through the machine learning with the, teacher data that is created by the provider. The information processing device 1 then calculates the predicted-reliability of each of the multiple pieces of second answer information to be outputted such that the piece of the second answer information that the user further seeks for has a high predicted-reliability, for example.
This enables the information processing device 1 to output second answer information in descending order of the calculated predicted-reliability (hereinafter, also referred to as priority), for example. Accordingly, the information processing device 1 enables the user to preferentially read answer information that the user seeks for. In other words, the result of evaluation of each piece of the second answer information allows the information processing device 1 to perform priority control for more preferentially presenting more likely pieces of the second answer information as answer information that the user seeks for.
[Hardware Configuration of Information Processing Device]
Next, a hardware configuration of the information processing device 1 will be described.
The information processing device 1 includes a CPU 101 that is a processor, a memory 102, an external interface (I/O unit) 103, and a storage medium 104. The respective units are connected to one another via a bus 105.
The storage medium 104 stores a program 110 for performing processing (hereinafter, also referred to as search control processing) of calculating the priority when the first answer information is outputted, in a program storage region (not illustrated) in the storage medium 104, for example. Moreover, the storage medium 104 includes an information storage region 130 (hereinafter, also referred to as storage unit 130) in which information used when the search control processing is performed is stored, for example.
As illustrated in
[Function of Information Processing Device]
Next, a function of the information processing device 1 will be described.
The CPU 101 of the information processing device 1 cooperates with the program 110 to operate as a keyword extracting unit 111 (hereinafter, also referred to as extracting unit 111 or receiving unit 111), a machine learning executing unit 112, an information receiving unit 113, and an information searching unit 114, for example. Moreover, the CPU 101 of the information processing device 1 cooperates with the program 110 to operate as a priority calculating unit 115 (hereinafter, also simply referred to as calculating unit 115), and a result outputting unit 116, for exampled. In addition, for example, teacher data 131, a viewpoint table 132, a calculation parameter 133, an identification function 134, and search target data 135 are stored in the information storage region 130. Note that, an explanation is hereinafter made by assuming that the teacher data 131 includes first question information 131a, first answer information 131b, and right/wrong information 131c which are associated with each other.
The keyword extracting unit 111 extracts keywords from the first question information 131a and the first answer information 131b which are included in the teacher data 131 stored in the information storage region 130. For example, the keyword extracting unit 111 extracts keywords by performing morpheme segmentation on the first question information 131a and the first answer information 131b.
Moreover, when the information searching unit 114 searches for second answer information 141b with keywords extracted from second question information 141a, the keyword extracting unit 111 extracts keywords from each of the second question information 141a and the second answer information 141b. For example, the keyword extracting unit 111 extracts keywords by performing morpheme segmentation on the second question information 141a and the second answer information 141b.
For example, the teacher data 131 includes information in which first question information 131a that the information processing device 1 highly likely receives, first answer information 131b that is an answer that the user seeks for, and the right/wrong information 131c indicating that the first answer information 131b is an appropriate answer to the first question information 131a are associated with each other. Moreover, for example, the teacher data 131 includes information in which the first question information 131a that the information processing device 1 highly likely receives, different first answer information 131b that is not an answer that the user seeks for, and the right/wrong information 131c indicating that the different first answer information 131b is not an appropriate answer to the first question information 131a are associated with each other.
This enables the information processing device 1 to execute machine learning for first question information while distinguishing first answer information that the user seeks for from first answer information (different first answer information) that the user does not seek for, as described later. A specific example of the teacher data 131 will be described later.
Meanwhile, the keyword extracting unit 111 may he configured to receive the input of the teacher data 131 when the provider or the like inputs the teacher data 131 to the information processing device 1.
The machine learning executing unit 112 executes machine learning on the calculation parameter 133 including parameter-values used for calculating a predicted-reliability of the first answer information 131b included in the teacher data 131, where the predicted-reliability indicates how much the first answer information 131b is likely to be an answer to the first question information 131a.
For example, the machine learning executing unit 112 specifies supplementary information (hereinafter, also referred to as first supplementary information, first correlation information, or first correlation degree) that is included in supplementary information associated with keywords extracted from the first question information 131a, and in supplementary information associated with keywords extracted from the first answer information 131b. The machine learning executing unit 112 then inputs the first supplementary information and the priority of the first answer information 131b, as learning data, to the identification function 134 so as to adjust the parameter-values included in the calculation parameter 133. The identification function 134 is a function for outputting a predicted-reliability of the first answer information 131b, its other words, a function for outputting priority of the first answer information 131b, when the first supplementary information and the calculation parameter 133 at inputted, for example. Further, when the right/wrong information 131c associated with the first answer information 131b indicates that the first answer information is an appropriate answer, the machine learning executing unit 112 may input, as learning data, “1.0” as the priority of the first answer information 131b to the identification function 134, for example. Meanwhile, when the right/wrong information 131c associated with the first answer information 131b indicates that the first answer information is an inappropriate answer, the machine learning executing unit 112 may input, as learning data, “0.0” as the priority of the fast answer information 131b to the identification function 134, for example. Further, the machine learning, executing unit 114 executes machine learning on the calculation parameter 133 for each piece of first supplementary information, for example.
In other words, every time learning data is inputted to the identification function 134, the machine learning executing unit 112 adjusts the calculation parameter 133 so that the identification function 134 is established not only for learning data inputted in the past but also for learning data newly inputted. This enables the machine learning executing unit 112 to improve the accuracy of the calculation parameter 133 every time learning data is inputted into the identification function 134. Accordingly, even when first supplementary information that is not subjected to machine learning is inputted, the priority calculating unit 115 is capable of predicting and outputting the priority of the lint answer information 131b associated with the inputted first supplementary information with the generalization function of the machine learning, as described later.
Note that, the machine learning executing unit 112 may operate in accordance with an algorithm, such as adaptive regularization of weight vectors (AROW), confidence weighted (CW), or sou confidence weighted learning (SCW).
The information receiving unit 113 receives new question information (hereinafter, also referred to as second question information 141a) transmitted by the provider terminal 11.
The information searching unit 114 searches for answer information (hereinafter, also referred to as second answer information 141b) to the second question information 141a by using keywords extracted by the keyword extracting unit 111. For example, the information searching unit 114 searches the search target data 135 including multiple pieces of answer information prepared in advance by the provider, for the second answer information 141b. The search target data 135 may include answer information the same as the first answer information 131b included in the teacher data 131. Further, the provider may utilize a search engine for open source as the information searching unit 114, for example.
Before multiple pieces of the second answer information 141b searched out by the information searching unit 114 with the second question information 141a are outputted, the priority calculating unit 115 calculates the priority of each of the multiple pieces of the second answer information 141b by using the calculation parameter 133 stored in the information storage region 130. For example, the priority calculating unit 115 specifies, for each of the multiple pieces of the second answer information 141b, supplementary information (hereinafter, also referred to as second supplementary information, second correlation information, or second correlation degree) that is included in supplementary information associated with keywords extracted from the second question information 141a, and also included in supplementary information associated with keywords extracted from the each piece of the second answer information 141b. The priority calculating unit 115 inputs the second supplementary information and the calculation parameter 133 to the identification function 134, and acquires the priority outputted as the priority of the second answer information 141b.
The result outputting unit 116 transmits the multiple pieces of the second answer information 141b searched out by the information searching unit 114 to the provider terminal 11. The provider terminal 11 then outputs the received multiple pieces of the second answer information 141b in descending order of the priorities (predicted-reliabilities) calculated by the priority calculating unit 115 to an output device (output device viewable by the user), for example. Note that, the viewpoint table 132 will be described later.
Next, a first embodiment will be described.
The information processing device 1 waits until machine learning execution timing comes (NO at S1). The machine learning execution timing is timing when the provider executes machine learning of the teacher data 131, for example. Specifically, the machine learning execution timing may be timing when the provider performs an input indicating that machine learning of the teacher data 131 is executed, for example.
When the, machine learning execution timing comes (YES at S1), as illustrated in
The information processing device 1 then specifies supplementary information associated with the keywords that were extracted in the processing at S2 (S4). Moreover, the information processing device 1 specifies supplementary information associated with the keywords that were extracted in the processing at S3 (S5). Thereafter, the information processing device 1 executes machine learning based on the supplementary information specified in the processing at S4, the supplementary information specified in the processing at S5, and the right/wrong information 131c indicating whether the first answer information 131b is a right answer to the first question information 131a (S6). In other words, the information processing device 1 executes machine learning for the first question information 131a while distinguishing the first answer information 131b that is answer information that the user seeks for from the first answer information 131b that is answer information that the user does not seek for.
Thereafter, the information processing device 1 waits until information output timing (NO at S11). The information output timing is tuning when the information processing device 1 searches for the second answer information 141b with the second question information 141a, for example. When the information output timings comes (YES at S11), as illustrated in
The information processing device 1 then specifies supplementary information associated with the keywords that were extracted in the processing at S12 (S14). Moreover, the information processing device 1 specifies supplementary information associated with the keywords that were extracted in the processing at S13 (S15). Thereafter, the information processing device 1 calculates the predicted-reliability (priority) of answer information included in each of the multiple pieces of the second answer information 141b, based on the supplementary information specified in the processing at S14 and the supplementary information specified in the processing at S15 (S16).
In other words, when the information processing device 1 searches for the second answer information 141b, the information processing device 1 refers to the calculation parameter 133 obtained in advance through the machine learning of information on the first answer information 131b that the user seeks for to the first question information 131a, and calculates the priority with which the second answer information 141b is to be outputted. The information processing device 1 then outputs second answer information in descending order of the calculated priorities, for example. This allows the information processing device 1 to preferentially output the second answer information 141b that the user seeks for.
In this manner, the information processing device 1 in this embodiment extracts keywords from the first question information 131a and the first answer information 131b, which are included in the teacher data 131. The information processing device 1 then executes machine learning on the calculation parameter 133 for calculating the predicted-reliability of the first answer information 131b which indicates how much the first answer information 131b is likely to he an answer that is responsive to the first question information 131a. Specifically, the information processing device 1 executes machine learning based on supplementary information associated with the keywords extracted from the first question information 131a, supplementary information associated with the keywords extracted from the first answer information 131b, and the right/wrong information 131c indicating whether the first answer information 131b is a right answer to the first question information 131a.
Thereafter, when the information processing device 1 outputs multiple pieces of the second answer information 141b in response to the inputted second question information 141a, the information processing device 1 calculates the predicted-reliabilities of the multiple pieces of the second answer information 141b. For example, the information processing device 1 calculates the predicted-reliability of each of the multiple pieces of the second answer information 141b with the calculation parameter 133 obtained through the machine learning, based on supplementary information associated with the keywords extracted from the second question information 141a and supplementary information associated with the keywords extracted from the each piece of the second answer information 141b.
This allows the information processing device 1 to preferentially output the second answer information 141b that the user is highly likely to seeks for. Accordingly, the information processing device 1 allows the user to preferentially read the second answer information 141b that the user is highly likely seeks for.
Next, the details of the first embodiment will be described.
As illustrated in
[Specific Example of Teacher Data]
In the example illustrated in
In other words, the teacher data 131 illustrated in
This allows the information processing device 1 to execute machine learning for the first question information 131a while distinguishing the first answer information 131b that the user seeks for from the first answer information 131b that the user does not seek for, as described later. An explanation of other information included in
[Specific Example of Keywords Extracted from Question Information and Answer Information]
Next, a specific example of keywords (hereinafter, also referred to as keyword information) extracted from the first question information 131a and the first answer information 131b will be described.
The keyword information illustrated in
For example, in the keyword information illustrated in
Referring back to
[Specific Example of Viewpoint Table]
For example, in the viewpoint table 132 illustrated in
[Specific Example of Supplementary Information Specified in Processing at S24]
Next, a specific example of supplementary information specified in the processing at S24 will be described.
The supplementary information illustrated in
For example, in “KEYWORD (QUESTION INFORMATION)” for a piece of information whose “ITEM NUMBER” is “1” in the keyword information illustrated in
Subsequently, for example, the machine learning executing unit 112 refers to the viewpoint table 132 illustrated in
In other words, the machine teaming executing unit 112 specifies “PRODUCT CATEGORY-AAA”, “PRODUCT NAME-AAA MANAGER”, and “PHASE-EXECUTION”, as supplementary information associated with a piece of information (the first question information 131a) set to “KEYWORD (QUESTION INFORMATION)” of a piece of information whose “ITEM NUMBER” is “1” in
Accordingly, as illustrated in a piece of information whose “ITEM NUMBER” is “1” in
Note that, in the viewpoint table 132 illustrated in
Specific Example of Supplementary Information Specified in the Processing at S25]
Next, a specific, example of supplementary information specified in the processing at S25 will be described.
The supplementary information illustrated in
Specifically, “ERROR MESSAGE”. “OCCURRENCE”, “PROCESS”, “NETWORK JOB”, “RUN”, “PROCESS”, and “STOP”, as keywords, are set to “KEYWORD (ANSWER INFORMATION)” of a piece of information whose “ITEM NUMBER” is “1” in the keyword information illustrated in
Subsequently, for example, the machine learning executing unit 112 refers to the viewpoint table 132 illustrated in
In addition, for example, the machine learning executing unit 112 refers to the viewpoint table 132 illustrated in
Moreover, for example, the machine learning executing unit 112 refers to the viewpoint table 132 illustrated in
Accordingly, as illustrated in a piece of information whose “ITEM NUMBER” is “1”
Further, as illustrated in a piece of information whose “ITEM NUMBER” is “4” in
Note that, in the viewpoint table 132 illustrated in
Referring back to
[Specific Example of First Supplementary Information Specified in Processing at S26]
For example, supplementary information that is included in common in the supplementary information explained in
Further, “1 (TIME)” is set to “COUNT” of a piece of information whose “SUPPLEMENTARY INFORMATION” is set at “PRODUCT CATEFORY-AAA” in
Similarly, as illustrated in
Further, as illustrated in
Referring back to
In other words, the machine learning, executing unit 112 specifies first supplementary information by comparing the supplementary information that is a higher-level concept of keywords extracted from the first question information 131a with the supplementary information that is a higher-level concept of keywords extracted from the first answer information 131b. Therefore, for example, when multiple keywords having the similar meaning but varying in style are included in the first question information 131a, the machine learning executing unit 112 is able to perform processing by regarding these keywords as the same supplementary information. Moreover, for example, when multiple keywords having the similar meanings but varying in style are present in both the first question, information 131a and the first answer information 131b, the machine learning executing unit 112 is also able to perform processing by regarding these keywords as the same supplementary information.
This allows the machine learning executing unit 112 to exclude a slight difference in expression and the like between keywords when executing machine learning of the keywords extracted from the first question information 131a and the first answer information 131b as learning data, as described later. This allows the machine learning executing unit 112 to execute machine learning so that the contents respectively included in the first question information 131a and the first answer information 131b are reflected more accurately.
For example, the machine learning executing unit 112 specifies first supplementary information that is included in both supplementary information associated with keywords extracted from the first question in 131a in the processing at S27 and the supplementary information associated with keywords extracted from the first answer information 131b. The machine learning executing unit 112 then inputs the first supplementary Information and the priority of the first answer information 131b as learning data to the identification function 134 so as to adjust the calculation parameter 133. In the case, the machine learning executing unit 112 executes machine learning on the calculation parameter 133 for each piece of first supplementary information, for example.
In other words, the machine learning executing unit 112 adjusts the calculation parameter 133 every time learning data is inputted to the identification function 134 so that the identification function 134 is established not only for learning data inputted in the past but also for learning data newly inputted. This allows the machine learning executing unit 112 to improve the accuracy of the calculation parameter 133 every time teaming data is inputted to the identification function 134. Accordingly, even when first supplementary information that is not subjected to machine learning is inputted, the priority calculating unit 115 is able to predict and output the priority of the first answer information 131b associated with the inputted first supplementary information with the generalization function of the machine learning, as described later. A specific example of the calculation parameter 133 will be described later.
Referring back to
Thereafter, the information searching unit 114 of the information processing device 1 executes a search for the second answer information 141b by using keywords extracted in the processing at S32 (S33). Hereinafter, specific examples of the second question information 141a and the second answer information 141b will be described.
[Specific Example of Second Question information Received in Processing at S31]
For example, in the second question information 141a illustrated in
[Specific Example of Second Answer Information Searched in Processing at S33]
Next, a specific example the second answer information 141b will be described.
For example, the second answer information 141b illustrated in
For example, in the second answer information 141b illustrated in
Referring back to
In other words, the keyword extracting unit 111 and the priority calculating unit 115 perform the processing at S32 and from S34 to S37, which is the same as the processing from S22 to S26 explained in
[Specific Example of Second Supplementary information Specified in Processing at S37]
For example, in the second supplementary in illustrated in
Referring back to
[Specific Example of Calculation Parameter]
For example, in the calculation parameter 133 illustrated in
[Specific Example of Priority Information]
For example, “PRODUCT CATEGORY-AAA”, “PRODUCT NAME-AAA MANAGER”, and “PHASE-EXECUTION” are set to “SUPPLEMENTARY INFORMATION” for the second supplementary information explained in
In other words, for example, the priority calculating unit 115 calculates priority so that the priority of the second answer information 141b, whose matching degree between the first supplementary information associated with the right/wrong information 131c indicating a right answer and the second supplementary information is higher than that of different second answer information 141b, becomes higher than the priority of the different second answer information 141b. Meanwhile, for example, the priority calculating unit 115 calculates priority so that the priority of the second answer information 141b, whose matching degree between the first supplementary information associated with the right/wrong information 131c indicating a wrong answer and the second supplementary information is higher than that of different second answer information 141b, becomes lower than the priority of the different second answer information 141b.
The priority calculating unit 115 then determines an output order of the pieces of second answer information 141b in descending, order of values set to “PRIORITY”, for example. Accordingly, as illustrated in
Referring hack to
This allows the information processing device 1 to preferentially output a piece of second answer information 141b that the user is highly likely to seek for. Accordingly, the information processing device 1 allows the user to preferentially read the piece of second answer information 141b that the user is highly likely to seek for.
Further, in the processing at S26, the machine learning executing unit 112 may specify first supplementary information by considering information other than the supplementary information specified at S24 and the supplementary information specified S25.
In this case, for example, the machine learning executing unit 112 causes the information searching unit 114 to execute a search for the first answer information 131b with keywords extracted from the first question information 131a. The machine learning executing unit 112 acquires, for each piece of first answer information 131b that is searched out with the keywords extracted from the first question information 131a, information (hereinafter, also referred to as search score) indicating the, priority of the output calculated by the information searching unit 114, for example. Thereafter, for example, the machine learning executing unit 112 sets the acquired search score as a part of first supplementary information. Hereinafter, a specific example of the first supplementary information to search be described.
[Specific Example of First Supplementary Information to Which Search Score is Set]
In this case, the priority calculating unit 115 acquires a search score of each piece of second answer information 141b that is searched out when the processing at S33 is executed. Moreover, the priority calculating unit 115 sets the acquired search score as a part of the second supplementary information. This allows the priority calculating unit 115 to determine the output priority of the second answer information 141 with higher accuracy in the processing at S41.
In this manner, the information processing device 1 in this embodiment extracts keywords from the first question information 131a and the first answer information 131b, which are included in the teacher data 131. The information processing device 1 then executes machine learning on the calculation parameter 133 for calculating the predicted-reliability of the first answer information 131b indicating how much the first answer information 131b is likely to be an answer that is responsive to the first question information 131a. For example, the information processing device 1 executes machine learning, based on supplementary information associated with keywords extracted from the first question information 131a, supplementary information associated with keywords extracted from the first answer information 131b, and the right/wrong information 131c indicating whether the first answer information 131b is a right answer to the first question information 131a.
Thereafter, when the information processing device 1 outputs multiple pieces of the second answer information 141b associated with the inputted second question information 141a, the information processing device 1 calculates the predicted reliabilities of the multiple pieces of the second answer information 141b. For example, the information processing device 1 calculates the predicted-reliability of each of the multiple pieces of second answer information 141b, with the calculation parameter 133 obtained through the machine learning, based on supplementary information associated with keywords extracted from the second question information 141a and supplementary information associated with the keywords extracted from the each piece of second answer information 141b.
This allows the information processing device 1 to preferentially output a piece of second answer information 141b that the user seeks for. Accordingly, the information processing device 1 allows the user to preferentially read the piece of second answer information 141b that the user seeks for.
Next, a second embodiment will be described.
The information processing device 1 in the first embodiment executes machine learning on the calculation parameter 133, and refers to the calculation parameter 133 obtained through the machine learning to determine the output priority of the second answer information 141b.
In contrast, the information processing device 1 in the second embodiment does not perform processing of executing the machine learning on the calculation parameter 133 (the processing from S21 to S27 explained in
For example, in the priority information illustrated in
With this, the information processing device 1 in the second embodiment does not have to execute machine learning on the calculation parameter 133. Moreover, the information processing device 1 in the second embodiment does not have to perform the input and the like to the identification function 134 when the total count is decided, so that the information processing device 1 is able to easily determine output order of the second answer information 141b.
All examples and conditional language recited herein are intended for pedagogical purposes to aid the reader in understanding the invention and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions, nor does the organization of such examples in the specification relate to a showing of the superiority and inferiority of the invention. Although the embodiments of the present invention have been described in detail, it should be understood that the various changes, substitutions, and alterations could be made hereto without departing from the spirit and scope of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2016-035461 | Feb 2016 | JP | national |
