Question and answer system, question and answer program, and question and answer method

Abstract

To prevent a questioner from being guided to the dangerous operation in a question answering system.SOLUTION: A question-answering pair replacement processing part includes: a dangerous operation determination part for determining whether the dangerous operation is present for question-answering pair data; and a response sentence replacement part for replacing the response sentence having description of the dangerous operation included in the document when it is determined that the dangerous operation is present, with a post-replacement response sentence corresponding to the classification of the dangerous operation and creating the replaced question-answering pair data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a question-and-answer system, a question-and-answer program, and a question-and-answer method.

Background Art

[0002] With the development of artificial intelligence technology and natural language processing technology, the use of question-and-answer systems via chatbots and the like has been expanding. Conventionally, it has been often used for casual conversations and simple responses, but in recent years, its use in product customer support in companies has been spreading.

[0003] The main purpose of using a question-and-answer system in product customer support is to quickly and accurately answer customer inquiries and thus solve customer problems at an early stage. The ability to solve customer problems at an early stage leads to a reduction in the response cost on the support side, which is beneficial to both the customer and the product-providing company.

[0004] In order to answer customer inquiries quickly and accurately, a question-and-answer system needs to be able to answer as many questions as possible. Most question-and-answer systems prepare assumed questions and response texts for them in advance, and answer the response texts for assumed questions similar to the inquiry content. Therefore, it is important to prepare a large number of pairs of assumed questions and response texts in order to answer customer inquiries quickly and accurately.

[0005] However, the preparation of these assumed questions and response texts generally requires a large amount of man-hours. Therefore, it is difficult to manually create a large number of assumed questions and response texts. Thus, a method of mechanically generating assumed questions and response texts from existing materials has been proposed.

[0006] For example, Patent Document 1 mechanically generates assumed questions and response texts by extracting existing FAQs (Frequently Asked Questions) in websites and manuals and converting them for a question-and-answer system.

[0007] Further, Patent Document 2 mechanically generates assumed question-and-answer sentences by converting descriptions following a specific pattern into assumed question-and-answer sentences from the sentences and document formats in the manual.

Prior Art Documents

Patent Documents

[0008]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0009] The methods of Patent Documents 1 and 2 can generate a large number of assumed question-and-answer sentences. On the other hand, since these methods mechanically generate response sentences, there is a possibility of guiding customers to perform dangerous operations that are disadvantageous to them.

[0010] For example, when the original material includes a description of a dangerous operation, such as an operation to stop the function of a product, the above method will guide it as it is as a response to the inquiry. Depending on the way the customer makes an inquiry, the response of the question-and-answer system may return an incorrect response to the customer's problem, even though it is correct as a response to the assumed question.

[0011] In this case, the customer may trust the response of the question-and-answer system and perform an incorrect operation on the customer's problem. This is contrary to the purpose of introducing the question-and-answer system, which is to answer the customer's inquiries accurately and quickly.

[0012] An object of the present invention is to prevent a questioner from being guided to perform a dangerous operation in a question-and-answer system.

Means for Solving the Problems

[0013] The question-and-answer system according to one aspect of the present invention identifies a question pattern and a response pattern corresponding to the question pattern from the description included in a document, and converts the identified question pattern and response pattern to create question-and-answer pair data including a question sentence and a response sentence. The question-and-answer system includes a question-and-answer pair generation processing unit and a question-and-answer pair replacement processing unit that replaces the question-and-answer pair data with replaced question-and-answer pair data including the question sentence and a replaced response sentence. The question-and-answer pair replacement processing unit includes a dangerous operation determination unit that determines whether there is a dangerous operation in the question-and-answer pair data, and when it is determined that the dangerous operation exists, a response sentence replacement unit that replaces the response sentence having the description of the dangerous operation included in the document with the replaced response sentence according to the classification of the dangerous operation to create the replaced question-and-answer pair data.

Effect of the Invention

[0014] According to one aspect of the present invention, in a question-and-answer system, it is possible to prevent guiding a questioner to perform a dangerous operation.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Modes for Carrying Out the Invention

Examples

[0016] FIG. 1 shows the overall diagram 100 of the question-and-answer system targeted by this Example 1. In the question-and-answer system, when a customer 140 wants to inquire with the device developer when a problem occurs in a device 150 owned by the customer, etc., the customer 140 exchanges questions and answers with the GUI 190 provided by the question-and-answer computer 110.

[0017] The question-and-answer computer 110 responds to the questions of the customer 140 using the replaced question-and-answer data 118. This replaced question-and-answer data 118 is created as follows. First, question-and-answer data 112 is created by extracting and formatting assumed question-and-answer pairs from the existing document 120 in the question-and-answer pair generation processing unit 111. The response sentences in the question-and-answer data 112 are replaced in the question-and-answer pair replacement processing unit 113. First, the dangerous operation determination unit 114 determines whether the question-and-answer pair includes a response corresponding to a dangerous operation, or uses the customer information 116 and device information 117 for the determination. If it is determined that there is a match, the subsequent response sentence replacement unit 115 replaces the response sentence with a safe sentence. In this way, the replaced question-and-answer pair data 118 is created.

[0018] The customer 140 and the device 150 may constantly notify their states to the question-and-answer system 110. In that case, the determination result of a dangerous operation changes due to the update of the customer information 116 and the device information 117.

[0019] Hereinafter, each element and the processing flow constituting the question-and-answer system will be described in detail.

[0020] FIG. 2 shows a configuration diagram of the question-and-answer computer 200. The question-and-answer computer 200 has a CPU 210, a memory 220, a network interface 240, and a display 250. The CPU 210 determines the operation of the question-and-answer computer 200 according to various programs stored in the memory 220. The memory 220 stores a question-and-answer pair generation program 221, a question-and-answer pair replacement program 222, a dangerous operation determination program 223, a sentence replacement program 224, a question-and-answer program 226, a document 230, question-and-answer pair data 231, replaced question-and-answer pair data 232, customer information 233, and device information 234.

[0021] The question-and-answer pair generation program 221 refers to the document 230 and creates question-and-answer pair data 231. As functions that make up the question-and-answer pair generation program 221, it has a structure analysis unit 240, a text analysis unit 250, and a response data generation unit 260. The structure analysis unit 240 includes a layout analysis unit 241, a chapter hierarchy analysis unit 242, a table format analysis unit 243, a figure format analysis unit 244, and the like. The structure analysis unit 240 is not limited to these and can include a program for analyzing the structure of a document.

[0022] The text analysis unit 250 has a morphological analysis unit 251, a dependency analysis unit 252, a reference analysis unit 253, a regular expression unit 254, and the like. In addition to these, the text analysis unit 250 can have processing units necessary for analyzing various natural languages. For example, if it is English, processing such as Stemming, and if it is Chinese, processing such as word decomposition can be mentioned. In addition, the text analysis unit 250 is not limited to these and can include other programs for analyzing text information. The response data generation unit 260 has a pattern database 261 and a synonym / paraphrase expansion unit 262.

[0023] The question-and-answer pair replacement program 222 refers to the question-and-answer pair data 231 and creates replaced question-and-answer pair data 232 through the processing of the dangerous operation determination program 223 and the text replacement program 224. The question-and-answer program 226 is a program for conducting question-and-answer exchanges with the customer 140 using the replaced question-and-answer pair data 232.

[0024] Document 230 is data used for generating question-and-answer pair data 231. Document 230 includes documents, files, and media that publish texts, drawings, and data. For example, it includes product manuals, catalogs, FAQ collections, websites, user forums, databases of past inquiries, databases of product information, etc. Question-and-answer pair data 231 and replaced question-and-answer pair data 232 are data that list assumed questions used in the question-and-answer system and response texts to those questions. The replaced question-and-answer pair data 232 is equivalent to the question-and-answer pair data 231 except that the content of the data has been partially changed by the question-and-answer pair replacement program 222. Customer information 233 is information about the customer 140 who makes the inquiry. Device information 234 is information about the device 150 held by the customer 140. The customer information 233 and the device information 234 are not essential in this embodiment.

[0025] The network interface 240 is used when communicating with other computers. For communication, protocols such as TCP / IP (Transmission Control Protocol / Internet Protocol), HTTP (HyperText Transfer), and SSH (Secure Shell) built on top of TCP / IP can be used. The display 250 is a device that displays a screen for the question-and-answer interaction with the customer 140. If there are other devices capable of question-and-answer interaction, the display 250 can be replaced by such a device.

[0026] For example, when conducting question-and-answer interaction by voice, it can be replaced by a microphone and a speaker. Also, the question-and-answer computer 200 itself may not have a display 250, and instead, information indicating the content of the screen display, such as HTML (HyperText Markup Language), is transmitted via the network interface 240, and the question-and-answer interaction is conducted on the display of the computer held by the customer who receives it.

[0027] The programs 221, 222, 223, 224, 225 and data 230, 231, 232, 233 that the question-and-answer computer 200 has do not all have to be in a single computer and may be divided among multiple computers. For example, the question-and-answer pair generation program 221, the question-and-answer pair replacement program 223, and the question-and-answer program 225 may each operate on a different computer. In this case, the question-and-answer pair data 231 generated by the computer having the question-and-answer pair generation program 221 is transmitted via the network interface 240 to the memory 220 in the computer having the question-and-answer pair replacement program 223. Similarly, the replaced question-and-answer pair data 232 generated from the question-and-answer pair data 231 received by the computer having the question-and-answer pair replacement program 223 is transmitted via the network interface 240 to the memory 220 in the computer having the question-and-answer program 225. The computer having the question-and-answer program 225 can be configured to perform question-and-answer exchanges using the received replaced question-and-answer pair data 232.

[0028] In FIG. 3, a configuration example of the question-and-answer pair data 300 used by the question-and-answer program 225 for question-and-answer is shown. In the example of FIG. 3, a question-and-answer pair assuming a storage device that includes a disk and stores data is shown. Therefore, the stored data, and the question sentences and answer sentences regarding the provided disk are arranged. The question-and-answer pair data 300 is a table that lists the correspondence between the question sentences 310 and the answer sentences 320 and stores the corresponding entries line by line. For example, in the question-and-answer pair data 300 shown in the figure, three pairs of question sentences and answer sentences are registered as entries 331, 332, and 333.

[0029] When the question-and-answer program 225 receives a question sentence input by a customer, it searches for an entry among entries 331, 332, and 333 of the question-and-answer pair data 300 that is close to the question sentence input by the customer. If an entry having a close question sentence exists, the answer sentence 320 of that entry is output as the answer of the question-and-answer program 225.

[0030] To determine the proximity of question sentences, various natural language processing techniques used in language processing can be utilized. For example, calculation methods such as the frequency of the same words appearing in both sentences, the BLEU (BiLingual Evaluation Understudy) value, and the distance between vectors using Word Embedding can be applied.

[0031] Figure 4 shows a configuration example of document 400. Document 400 is a hierarchical document, and patents, papers, manuals, reports, etc. are applicable.

[0032] Document 400 has a structure by arranging data such as text, figures, and tables according to some hierarchy or layout. This structure is defined by the position, content, size, decoration of the text, and the fact that they are separated by ruled lines. In the example of the figure, it can be considered that document 400 represents one chapter with title 410, and there are three sections indicated by section titles 420, 430, and 440 in that chapter.

[0033] In the section corresponding to section title 420, after the section text 421, bullet points 422 are arranged. Similarly, in the section corresponding to section title 430, after the section text 431, bullet points 432 are arranged. In the section corresponding to section title 440, after the section text 441, a table caption 442 and a table 443 are arranged. That is, this document 400 shows a hierarchical structure where sections come after the chapter and section text comes after the section.

[0034] FIG. 5 shows an example of the structure information 500 analyzed from the structure of the document 400 and represented in the form of a tree structure. The structure information 500 is represented as a tree structure formed by a group of nodes with the root node 505 as the root. In this structure information 500, the relationships of inclusion in the document are represented as parent-child relationships. For example, the root node 505 has a node 510 corresponding to a chapter as its child, and the node 510 corresponding to the chapter has nodes 520, 530, and 540 corresponding to sections. The nodes 520, 530, and 540 corresponding to the sections have, related to the content of the sections, nodes 521, 531, and 541 corresponding to the section text, nodes 522 and 532 corresponding to bullet points, a node 542 corresponding to a table, etc. as their children. The nodes 522 and 532 corresponding to the bullet points each have nodes 523, 524 and nodes 533, 534 corresponding to the respective items constituting the bullet points.

[0035] The node 542 corresponding to the table has nodes 543, 546, and 549 corresponding to the respective rows constituting the table, and the nodes 543, 546, and 549 corresponding to the rows have nodes 544, 545, 547, 548, 550, and 551 corresponding to the respective cells constituting the rows. The table may take different representation methods in the structure information. For example, nodes corresponding to the columns constituting the table may be used as child nodes of the node corresponding to the table, and the nodes corresponding to the columns may have nodes corresponding to the respective cells constituting the columns as their child nodes. Also, regardless of the order of columns and rows, all the cells constituting the table may be enumerated as nodes corresponding to the table and used as child nodes of the table.

[0036] Each node can similarly hold not only the hierarchical name (such as chapter, section, table, etc.) but also the text included in that part and information based on the structure (page number in the document, numbers of chapters, sections, tables, position of the text, font information) for the part of the document corresponding to the node.

[0037] In the first embodiment, in the tree structure shown in the structure information 500, a description that matches a pre-defined pattern, that is, a subtree of the tree structure is extracted, and a response sentence is generated. The pattern of this subtree of the tree structure is stored in the pattern database 261.

[0038] Figure 6 shows a pattern example 600. The pattern example 600 consists of three patterns 610, 611, and 612. The patterns 610, 611, and 612 consist of an extraction pattern 620 corresponding to a part of the tree structure of the structure information, and a response data template 630 that is the source of the question-and-answer pair generated when a description matching the pattern is extracted.

[0039] The extraction pattern description 621 shows an example of the extraction pattern 620. In this pattern, the structure to be extracted is shown by listing the hierarchical names and texts of the nodes in a parent-child relationship in the tree structure in pairs. This example shows the case where the hierarchical name 622 "section" and the hierarchical name 624 "section text" are in a parent-child relationship. Also, slots 623 "<phrase>" and 625 "<meaning>" are described corresponding to each hierarchical name. This indicates that in the extracted structure, the text of the corresponding node is substituted into these slots.

[0040] The extraction pattern description 641 shows another example of the extraction pattern 620. The extraction pattern description 641 shows a pattern that matches a tabular structure using a plurality of hierarchical names 642, 643, and 645.

[0041] The extraction pattern description 661 shows another example of the extraction pattern 620. The extraction pattern description 661 does not define a plurality of hierarchical names, and a sentence pattern 664 having slots 662 and 663 is described. This means that this extraction pattern description 661 matches any node regardless of the hierarchical name in the hierarchical structure. On the other hand, it is required that the node has a sentence that matches the sentence pattern 664.

[0042] As a method of describing the extraction pattern 620, in addition to the method shown in FIG. 6, a technique of flexibly establishing a correspondence relationship between tree structures can also be incorporated. For example, in the paper "Taxonomy of XML schema languages using formal language theory. ACM Trans. Internet Technol. 5, 4 (November 2005), 660-704.", a method for flexibly extracting a partial tree that matches a pattern from a document with a tree structure described in XML (Extensible Markup Language) is proposed.

[0043] The response data template 630 is described as a pair of a question sentence and a response sentence. These question sentences and response sentences can include slots that appear in the extraction pattern 620 in the text. In this case, in the extracted partial tree, if there is text associated with the slot in the extraction pattern 620, the text is substituted into the slot in the response sentence to generate the response sentence.

[0044] Although not described in FIG. 6, a description for processing the output method of the slot may be added to the response data template 630. For example, in the case of Japanese, processing such as changing to an appropriate conjugation system, or in the case of English, appropriately changing the tense of the verb or the singular / plural form of the noun can be considered.

[0045] Note that although the document structure is represented as a tree structure in FIGS. 5 and 6, the same applies if another representation form can represent the partial structure. For example, a table in a document may be represented in the form of a multi-dimensional array instead of a tree structure.

[0046] FIG. 7 shows a question and answer pair data generation flow 700 in which a question and answer pair generation program 221 in the question and answer computer 200 generates question and answer pair data 112 from the document 120.

[0047] In step 710, the structure analysis unit 240 and the internal layout analysis unit 241, chapter hierarchy analysis unit 242, table format analysis unit 243, figure format analysis unit 244, etc. analyze the document 1200 and convert it into a tree-structured representation such as the hierarchy information 500. Existing technologies can be used for this conversion. For example, as a method of dividing a document file in a format that does not hold information about paragraphs, which corresponds to the layout analysis unit 241, into paragraphs, there is a method of regarding sentences located in the vicinity of each other as the same paragraph.

[0048] In step 720, for the tree-structured representation of the document 120 converted in step 710, the text information held by each node is analyzed. For this, the morphological analysis unit 251, dependency analysis unit 252, anaphoric analysis unit 253, etc. included in the text analysis unit 380 perform their respective processes.

[0049] In step 730, for each pattern stored in the pattern database 261, a subtree that matches the extraction pattern 620 is extracted from the tree-structured representation of the document 400. For extracting a group of nodes such that the relationships between the nodes match, the methods described in the aforementioned paper, etc. can be used. Further, the text at each node of the extracted subtree is collated with the text and slots in the extraction pattern 620 to determine whether a correspondence can be established. If no correspondence can be established, that subtree is regarded as not extractable. Regular expressions, etc. can be used for this collation process.

[0050] In step 740, for the subtree of the tree structure representation of the document 400 corresponding to the extraction pattern 620 to be implemented, the slots in the response data template 630 are filled, and response data is output. At this time, not only one response data is output from one subtree according to the response data template 630, but also a plurality of data may be output. For example, the response data obtained by replacing the words in the response data with synonyms or changing the word order by the synonym / paraphrase expansion unit 262 can be output together. Further, when there is a notation in the response data that refers to an item in the document 400, such as "Table 2" or "Page 30", after adding the content of the reference destination to the response sentence, it may be changed to a response sentence with the reference destination added, such as "the following table" or "the following description".

[0051] FIG. 8 shows an example of the response data 800 generated by implementing the response data generation flow 700 using the pattern example 600 from the document 400 and the corresponding hierarchical structure 500.

[0052] The entries 831, 832, and 833 of the response data are examples generated as a result of the nodes 520, 530, and 540 corresponding to the sections and their child nodes being associated with the pattern 621. In all cases, the content of the nodes 520, 530, and 540 corresponding to the sections is incorporated into the question sentence, and the content of the child nodes is incorporated into the response sentence. The response sentence of the entry 833 includes a table. This is the result of the additional processing of the content of the reference destination performed in step 740 because the reference destination of the description "Table 2" included in the node 541 is the node 542, which is Table 433 in the document 400.

[0053] The entries 834 and 835 of the response data are examples generated as a result of the nodes 546 and 549 corresponding to the data rows in the table and their child nodes being associated with the pattern 641. The entry 836 of the response data is an example generated as a result of the first sentence of the node 521 corresponding to the section text being associated with the pattern 661.

[0054] The assumed user 810 indicates the assumed user of the document 400 that is the source of the question-and-answer data. This information is set based on the attributes of the document 400 (for example, if it is a manual for maintenance staff, assume the product vendor's maintenance staff as the assumed user) or the descriptions in the document 400 (if the content is described in a chapter for user administrators, assume the user administrator as the assumed user).

[0055] As shown in FIG. 8, by applying the question-and-answer pair data generation flow 700 to the document 400, response data 800 can be generated. Here, since the response data 80 is mechanically generated from the document 400, the response text may include a response that guides a seemingly dangerous operation. For example, entry 832 prompts an operation of "deleting the volume". This is an operation that makes the important data in the volume unavailable when there is important data in the volume, so it is generally considered a dangerous operation.

[0056] FIG. 9 shows a response text replacement processing flow 900 that replaces the responses of such dangerous operations from the response data 800. The response text replacement processing flow 900 is implemented by the question-and-answer pair replacement program 222. The response text replacement processing flow 900 may be executed in advance before the customer 140 asks a question, or may be executed when the customer 140 asks a question. There are pros and cons to either approach. When executed in advance, since the question-and-answer pair data after the replacement process can be confirmed by a person before showing it to the customer 140, there is room to confirm the quality and content of the response text. On the other hand, when executed when the customer 140 asks a question, the replacement method can be changed according to the customer 140. However, in any case, since the question-and-answer program 226 uses the result of the response text replacement processing flow 900 for the response to the customer, it is necessary to complete the response text replacement processing flow 900 by the time of the response to the customer.

[0057] In step 910 of the response text replacement processing flow 900, first, a dangerous operation database 225 listing dangerous operations is created.

[0058] Figure 10 shows a dangerous operation database example 1000. The dangerous operation database example 1000 is represented as columns of entries 1051, 1052, and 1053 composed of classification 1010, expression 1020, and replacement method 1030. The classification 1010 indicates the classification of dangerous operations. The expression 1020 enumerates the expressions belonging to the classification 1010. The expression 1020 may be described in natural language and can include other expressions for instructing operations, such as commands or the names of operation buttons.

[0059] The replacement method 1030 describes how to replace the response text of the question-and-answer pair data 112 when the expression 1020 classified into the classification 1010 is included in the question-and-answer pair data 112. For example, in entry 1051, when the question-and-answer pair data 112 includes the expression "Delete data", since the response text causes data loss, it means replacing the text with a communication to the support department. As in entry 1053, it may be instructed not to perform this response (that is, instead, return the response data of other questions). The dangerous operation database 225 may be created manually for individual entries or may be created by extracting operations corresponding to a list of past problems and failures.

[0060] In step 920 of the response text replacement processing flow 900, it is determined whether each entry of the question-and-answer pair data 800 includes a dangerous operation. Specifically, if there is a description corresponding to the expression 1020 by referring to the dangerous operation database 225 in the entry, it is determined that the entry includes a dangerous operation corresponding to the classification 1010. In step 930, according to the determination result in step 920, the response text is replaced according to the dangerous operation database 225. If a single entry corresponds to multiple dangerous operation classifications, replacements corresponding to both classifications may be performed, or the dangerous operation classifications may be ranked and only the replacement related to the operation with the highest rank may be performed.

[0061] Figure 11 shows a replaced question-and-answer pair data example 1100. The replaced question-and-answer pair data example 1100 shows the result of replacing the question-and-answer pair data example 800 based on the dangerous operation database example 1000. The replaced question-and-answer pair data example 1100 is the same as the question-and-answer pair data example 800 in that it has the question sentence 410. However, it is different in that the answer sentence 420 has become the replaced answer sentence 1120 and the dangerous operation 1100 has been added.

[0062] Entries 1131, 1132, 1133, 1134, 1135, and 1136 are the results of applying the answer sentence replacement processing flow 900 to entries 831, 832, 833, 834, 835, and 836 respectively. Entry 831 includes the description "Make the file read-only" in the answer sentence 420, and this description is an expression classified as "data operation" in the dangerous operation database example 1000. Therefore, "data operation" is stored as the dangerous operation 1100 in entry 1131, and the answer sentence 1120 is given the sentence "The following operation will modify the data. Please proceed with caution." based on the instruction of the replacement method 1030 of entry 1052.

[0063] Similarly, entry 832 includes the description "Delete the volume" in the answer sentence 420, and this description is an expression classified as "data loss" in the dangerous operation database example 1000. Therefore, "data loss" is stored as the dangerous operation 1100 in entry 1132, and the answer sentence 1120 is replaced with the sentence "Please contact the support department." based on the instruction of the replacement method 1030 of entry 1051. Since entries 833, 834, 835, and 836 do not have descriptions corresponding to the expression 1020 of the dangerous operation database example 1000, there is no corresponding item for the dangerous operation 1110 in entries 1133, 1134, 1135, and 1136, and there is no change between the answer sentence 820 and the replaced answer sentence 1120.

[0064] By the response text replacement process flow 900, the replaced question-and-answer pair data 118 is generated from the question-and-answer pair data 112. Thereafter, the question-and-answer program 226 can use the replaced question-and-answer pair data 118 to conduct question-and-answer sessions with the customer 140. Thereby, the question-and-answer program 226 can avoid responses that prompt dangerous operations such as "volume deletion". Also, the question-and-answer program 226 may use the content of the dangerous operation 1100 in the replaced question-and-answer pair data 118 for response selection. For example, an entry 1132 including data loss in the dangerous operation 1100 may be judged not to be selected as a response target in the first place.

[0065] According to the first embodiment, using existing documents, it is possible to generate the question-and-answer pair data used by the question-and-answer system. Also, when there is an expression in the response text that prompts a dangerous operation, by replacing the response text or adding a warning text, it is possible to prevent the customer, who is the questioner, from receiving a response that prompts such an operation. Due to both effects, the question-and-answer system can answer more questions, and a reliable question-and-answer system that does not respond to dangerous operations can be realized.

Embodiment

[0066] When there are a large number of documents 120, the expressions used in the documents are also diverse. In that case, the labor required to comprehensively create the dangerous operation database 225 increases according to the amount of the documents. In the second embodiment, a method for assisting in creating the dangerous operation database 225 will be described.

[0067] In many cases, dangerous operations can be represented by a combination of the operation target and the operation content. Therefore, by considering the operation target and content separately, it is possible to easily determine whether various operations are dangerous.

[0068] FIG. 12 shows an example of the dangerous operation determination table 1200. The risk operation determination table 1200 is a two-axis table that obtains the classification 1010 of risk operations from the operation content 1210 and the operation target 1220 for the expressions representing the operations. In the risk operation determination table 1200, as the classification of the operation content, row entries 1211, 1212, 1213, 1214, and 1215 are defined. The classification 1230 of the operation content indicates the classification to which each entry belongs. Example 1231 gives examples of the expressions belonging to each classification. Generally, the verbs or nouns indicating the content of the operation are applicable.

[0069] For example, the row entry 1211 indicates that expressions such as "delete", "erase", and "exclude" are grouped into the classification of "delete". Similarly, as the classification of the operation target, column entries 1221, 1222, and 1223 are defined. The classification 1240 of the operation target indicates the classification to which each column entry belongs. Example 1241 gives examples of the expressions belonging to each classification. Generally, the nouns indicating the target of the operation are applicable. For example, the entry 1221 indicates that expressions such as "volume", "disk", and "file" are grouped into the classification of "data".

[0070] When there is an expression including a certain operation content and operation target, by using the risk operation determination table 1200 to obtain the corresponding row entry from the operation content and the corresponding column entry from the operation target, the classification 1010 of the risk operation in that row can be determined by referring to the cell 1250 where both entries intersect. Taking the example of the entry 1051, for the expression "delete the volume", since "delete" corresponds to the row entry 1211 and "volume" corresponds to the column entry 1221, it can be seen that by referring to the intersecting cell 1251, this expression is classified as "data loss".

[0071] According to this second embodiment, when creating the risk operation database 225 for a document including a large number of expressions, by classifying the operation content and the operation target individually and creating in advance a classification table of risk operations according to the combination of the classifications, the classification of a large number of risk operations becomes easy.

Embodiment

[0072] In the foregoing embodiment, in the response sentence substitution process flow 900, it was determined whether the question-and-answer pair included a dangerous operation based on the text of the question-and-answer pair data 800.

[0073] As another method for determining whether a dangerous operation is included, it is possible to use whether the assumed user of the document 112 from which the question-and-answer pair is generated matches the customer 140. For example, the assumed users of a document of a certain product may include end-users, administrator users, product vendor maintainers, etc. Generally, it is considered that administrator users have greater operation authority than end-users, and product vendor maintainers have greater operation authority than administrator users. For example, in the case of a storage device, an end-user does not have the authority to turn off the power, but an administrator user does. An end-user and an administrator user cannot refer to the operation log information, but a product vendor maintainer can refer to the operation log information. In that case, the content written in the document assuming the product vendor maintainer as the user should not be executed by the end-user or the administrator user. That is, a customer who is not a product vendor maintainer should not unconditionally respond to a document assuming the product vendor maintainer as the user and a question-and-answer pair generated from that document.

[0074] Therefore, a determination of dangerous operations is made using the information of the assumed user of the document. In making this determination, for each entry of the question-and-answer pair in the question-and-answer pair data 800 created by the question-and-answer pair data generation flow 700, the assumed user of the document 120 of the generation source is noted as the assumed user 840 for each entry.

[0075] FIG. 14 shows an example of a dangerous operation database 1400 in the third embodiment. The dangerous operation database example 1400 is represented as a column of entries 1451·1452 composed of classification 1010, condition 1420, and replacement method 1030. The classification 1010 and the replacement method 1030 are the same as those in the dangerous operation database example 1000 in Example 1. The condition 1420 consists of the assumed user 1421 of the document and the user attribute 1422. These indicate that when the assumed user of the question-and-answer pair and the attributes of the customer 140, who is the user of the question-and-answer system, match each entry, the question-and-answer pair is classified into the classification 1010 and replacement is performed according to the replacement method 1030.

[0076] Figure 15 shows the response text replacement flow 1500 in this Example 3. The response text replacement flow 1500 may be executed in advance before the customer 140 asks a question, or may be executed when the customer 140 asks a question. In step 1510 of the response text replacement flow 1500, for each entry in the question-and-answer pair data example 800, by collating the assumed user 840 and the attributes of the customer 140 with the dangerous operation database example 1400, the presence or absence of a dangerous operation is determined. In the dangerous operation database example 1400, if an entry where the condition 1420 matches is found, the corresponding entry in the question-and-answer pair data example 800 is determined to include the dangerous operation indicated by the classification 1010. In step 1520, if it is determined in step 1510 that the entry in the question-and-answer pair data example 800 includes a dangerous operation, it proceeds to step 930. If it is determined that it does not include a dangerous operation, it proceeds to step 910. Steps 910·920·930 perform the same processing as the response text replacement flow 900 in Example 1.

[0077] According to this Example 3, by using the information of the assumed user of the document to perform dangerous operation determination and response text replacement for the question-and-answer pair, it is possible to surely replace a response that prompts an operation that the customer 140 should not perform.

Example

[0078] In the foregoing embodiments, the determination of dangerous operations was defined by the content described in the document and the target users.

[0079] As another method for determining dangerous operations, in a situation where the questioner asks questions about the owned devices, there is a method based on the information of the devices owned by the questioner.

[0080] FIG. 13 shows an example of device information 1300. The device information 1300 has, as items, an owner 1310, a device ID 1311, a part / index 1312, and a status 1313. Also, the device information 1300 has a plurality of entries 1320 to 1329. Each entry describes the status of one part / index of a certain device. The owner 1310 indicates the customer who owns the device. The device ID 1311 indicates a unique character string for identifying the device. The part / index 1312 and the status 1313 indicate the parts that make up the device or the indices related to the device, and their statuses. The part may have a form or may be something without a form such as a software component. In the case of an index, items that can be evaluated as values such as performance values and operating times are described. The status may take two values of normal / abnormal or may be expressed as a quantitative value as in entries 1323 and 1327.

[0081] FIG. 16 shows a response text substitution process flow 1600 in the present Example 4. In the response text substitution process flow 1600, after step 920 in the response text substitution process flow 900, step 1610 is performed. In step 1610, for the entry of the question-response pair data example 800, if it has been determined to be a dangerous operation in step 920, the process proceeds to step 930. If it is not determined to be a dangerous operation, the process proceeds to step 1620. In step 1620, using the device information 1300, a determination of dangerous operation for the entry of the question-response data 800 is made.

[0082] Assume that the user attributes of the customer who is answering questions can be identified from previous questions and the like. Also, the device being questioned may be identified. In that case, the dangerous operation determination program 223 compares the user attributes and (if available) the device being questioned with the owner 1310 and device ID 1311 of the device information 1300, and extracts the entries where the owner and device match the question target. For example, if the questioner is "Alice" and the device is "ABC123", the extracted entries will be 1320·1321·1322·1323.

[0083] At this time, two procedures can be considered for the response text. The first is to first determine that the entries in the question and answer data 800 for devices other than the device "ABC123" are dangerous operations. By doing so, it prevents responding without modification to operations related to devices other than "ABC123".

[0084] The other is to prevent responding without modification to operations related to normal parts and indicators. Among the question and answer data 800, when the response text 420 contains an operation that affects the part or indicator 1312, and the state 1313 corresponding to that part or indicator 1312 contains a normal or quantitatively normal value, it means an operation is being applied to a normal part or indicator, so it can be determined as a dangerous response. Or, if the result of the operation is predictable and it can be predicted that the state 1313 will become abnormal (for example, an operation to turn off the power can be predicted to make the power in an abnormal state), it can also be determined as a dangerous response. After the dangerous operation determination in step 1620, proceed to step 930 to perform text substitution.

[0085] According to this Example 4, the question and answer system can suppress responses that prompt operations on devices and parts / indicators that are operating normally in response to inquiries from the questioner. Thereby, it can prevent the questioner from performing operations that cause abnormalities on devices and parts / indicators that are operating normally.

Example

[0086] The foregoing embodiments all showed a method for determining a dangerous operation at the time until the question-and-answer system returns a response.

[0087] In the fifth embodiment, a method is shown in which question-and-answer data is once used by the question-and-answer system, and then, in response to feedback from the questioner 140, dangerous operation determination and text replacement are performed.

[0088] In the fifth embodiment, the question-and-answer program 226 uses the question-and-answer pair data 112 before text replacement or the replaced question-and-answer pair data 118 to respond to an inquiry from the questioner 140.

[0089] When the questioner 140 who has received the response determines that the response includes a dangerous operation and contacts the question-and-answer system with the determination result, in the replaced question-and-answer pair data 118, the dangerous operation 1110 of the corresponding question-and-answer entry is updated according to the contact, and text replacement similar to step 930 is performed on the response text 1120 after replacement.

[0090] Multiple methods are conceivable for updating the dangerous operation 1110 upon receiving contact from the questioner 140. One is to directly contact the support 130 and convey that the response text includes a dangerous operation. In that case, the dangerous operation 1110 is updated according to the operation of the support staff 130. In another method, in the question-and-answer system, the questioner 140 is asked about their feelings regarding the response in text or in a selection format, and the dangerous operation 1110 is updated according to the result. In yet another method, a method using the device information 234 is conceivable. Suppose that as a result of the question-and-answer, the device information 234 is updated by the operation of the questioner 140, and the question-and-answer system recognizes that an abnormality has occurred in a specific part. In that case, it can be determined that the response text includes a dangerous operation.

[0091] According to the fifth embodiment, the question-and-answer system can identify and replace the response text including dangerous operations based on the feedback from the questioner and their owned devices in the question-and-answer process. Thereby, in subsequent question-and-answer sessions, it is possible to prevent the questioner from receiving responses including dangerous operations.

[0092] For each assumed question-response text pair, the question-and-answer system in the above embodiment determines whether it is an assumed question-response text pair that guides dangerous operations. If it is determined that it guides dangerous operations, the question-and-answer system does not use the response text as it is, but replaces it with a safe text and then responds.

[0093] According to the above embodiment, the question-and-answer system prevents guiding the questioner to perform dangerous operations. As a result, it prevents the questioner from performing incorrect operations according to the response.

Description of Reference Numerals

[0094] 100 Question-and-Answer System 110 Computer for Question-and-Answer 111 Question-Answer Pair Generation Processing Unit 112 Question-Answer Pair Data 113 Question-Answer Pair Replacement Processing Unit 114 Dangerous Operation Judgment Unit 115 Response Text Replacement Unit 116 Customer Information 117 Device Information 118 Replaced Question-Answer Pair Data 119 GUI 120 Document 130 Support Staff 140 Customer 150 Device

Claims

1. A question-and-answer system having a question-and-answer pair generation processing unit that identifies a question pattern and a response pattern corresponding to the question pattern from the descriptions included in a document, and creates question-and-answer pair data including a question sentence and a response sentence by converting the identified question pattern and response pattern, and a question-and-answer pair replacement processing unit that replaces the question-and-answer pair data with replacement question-and-answer pair data including the question sentence and a response sentence after replacement, wherein the question-and-answer pair replacement processing unit, has a dangerous operation determination unit that determines whether there is a dangerous operation in the question-and-answer pair data, and when it is determined that there is a dangerous operation, a response sentence replacement unit that replaces the response sentence having the description of the dangerous operation included in the document with the response sentence after replacement according to the classification of the dangerous operation, and creates the replacement question-and-answer pair data, characterized in that it has the above.

2. The dangerous operation determination unit, determines whether the response sentence included in the question-and-answer pair data includes a response having a description of the dangerous operation, and the response sentence replacement unit, when it is determined that the response sentence includes a response having a description of the dangerous operation, replaces the response sentence with the response sentence after replacement according to the classification of the dangerous operation, according to Claim 1 of the question-and-answer system described above.

3. The response sentence replacement unit, when it is determined that the response sentence includes a response having a description of the dangerous operation, according to Claim 2 of the question-and-answer system described above, is characterized in that a predetermined warning sentence is added to the response sentence according to the classification of the dangerous operation.

4. The response sentence replacement unit, when it is determined that the response sentence includes a response having a description of the dangerous operation, according to Claim 2 of the question-and-answer system described above, is characterized in that the text of the response sentence is replaced with another text different from the response sentence according to the classification of the dangerous operation.

5. The response sentence replacement unit, pre-creates a dangerous operation database including the classification of the dangerous operation, the expressions belonging to the classification of the dangerous operation, and the method of replacing the response sentence, and when it is determined that the response sentence includes a response having a description of the dangerous operation, according to Claim 2 of the question-and-answer system described above, is characterized in that the response sentence is replaced with the response sentence after replacement according to the classification of the dangerous operation by referring to the dangerous operation database.

6. The dangerous operation determination unit, The question-and-answer system according to claim 5, characterized in that, when referring to the dangerous operation database and there is an expression belonging to the classification of the dangerous operation in the response text, it is determined that the response includes the description of the dangerous operation.

7. The response text replacement unit The question-and-answer system according to claim 5, characterized in that the dangerous operation database is created by obtaining the classification of the dangerous operation from the operation content and the operation target.

8. The question-and-answer pair generation processing unit creates the question-and-answer pair data further including the assumed user of the document, The dangerous operation determination unit When determining whether the dangerous operation exists in the question-and-answer pair data, it is determined whether the dangerous operation exists or whether the response text has the description of the dangerous operation included in the document based on the relationship between the assumed user and the user, The response text replacement unit When it is determined that the dangerous operation exists based on the relationship between the assumed user and the user, the response text is replaced with the replaced response text according to the classification of the dangerous operation, The question-and-answer system according to claim 1, characterized in that when it is determined that the response text has the description of the dangerous operation included in the document, the response text having the description of the dangerous operation included in the document is replaced with the replaced response text according to the classification of the dangerous operation.

9. The dangerous operation determination unit The question-and-answer system according to claim 8, characterized in that when the assumed user and the user do not match, it is determined that the dangerous operation exists.

10. The dangerous operation determination unit When determining whether the dangerous operation exists in the question-and-answer pair data, it is determined whether the dangerous operation exists or whether the response text has the description of the dangerous operation included in the document based on the device information of the device owned by the user, The response text replacement unit When it is determined that the dangerous operation exists based on the device information, the response text is replaced with the replaced response text according to the classification of the dangerous operation, The question-and-answer system according to claim 1, characterized in that when it is determined that the response text has the description of the dangerous operation included in the document, the response text having the description of the dangerous operation included in the document is replaced with the replaced response text according to the classification of the dangerous operation.

11. The machine information includes information regarding the state corresponding to the machine. The dangerous operation determination unit The question-and-answer system according to claim 10, wherein the dangerous operation determination unit determines whether there is a dangerous operation according to whether the state is normal or abnormal.

12. The response sentence replacement unit The question-and-answer system according to claim 1, wherein the response sentence replacement unit replaces the response sentence with the replaced response sentence according to the classification of the dangerous operation based on the feedback information from the user.

13. A question-and-answer program that causes a computer to execute a question-and-answer pair generation step of identifying a question pattern and a response pattern corresponding to the question pattern from the descriptions included in a document, converting the identified question pattern and response pattern, and creating question-and-answer pair data including a question sentence and a response sentence, and a question-and-answer pair replacement step of replacing the question-and-answer pair data with replaced question-and-answer pair data including the question sentence and the replaced response sentence, The question-and-answer pair replacement step includes a dangerous operation determination step of determining whether there is a dangerous operation for the question-and-answer pair data, and, when it is determined that there is a dangerous operation, a response sentence replacement step of replacing the response sentence having the description of the dangerous operation included in the document with the replaced response sentence according to the classification of the dangerous operation, and creating the replaced question-and-answer pair data. The question-and-answer program is characterized by having the above.

14. A question-and-answer method including a question-and-answer pair generation step of identifying a question pattern and a response pattern corresponding to the question pattern from the descriptions included in a document, converting the identified question pattern and response pattern, and creating question-and-answer pair data including a question sentence and a response sentence, and a question-and-answer pair replacement step of replacing the question-and-answer pair data with replaced question-and-answer pair data including the question sentence and the replaced response sentence, The question-and-answer pair replacement step includes a dangerous operation determination step of determining whether there is a dangerous operation for the question-and-answer pair data, and, when it is determined that there is a dangerous operation, a response sentence replacement step of replacing the response sentence having the description of the dangerous operation included in the document with the replaced response sentence according to the classification of the dangerous operation, and creating the replaced question-and-answer pair data. The question-and-answer method is characterized by having the above.

Images