JP5485236B2 - FAQ creation support system and program - Google Patents

Description

  Embodiments described herein relate generally to FAQ creation support.

  Conventionally, FAQ (Frequently Asked Questions) that summarizes inquiries and their answers has been created and used as a communication tool for inquiring users and as a business tool for information providers who answer inquiries.

  In general, the FAQ analyzes a large number of query contents and answers to them, for example, a FAQ creator extracts a query with a high query frequency, or creates a representative query from queries with a high query frequency, Created by combining the extracted or created query with the appropriate answer.

JP 2009-15769 A

  It is an object of the present invention to provide a FAQ creation support system and program for realizing a FAQ creation environment that quantitatively evaluates a query and answer pair that is a candidate for a FAQ from a document group including a query sentence and an answer sentence.

The FAQ creation support system of the embodiment is based on the same query representative sentence among a plurality of query representative sentences extracted from the query sentences of each document in the document set of documents including the query sentence and the answer sentence. A first storage unit that stores a query representative sentence in which a document corresponding to a plurality of source query sentences is associated with the query representative sentence; and a plurality of answer representative sentences extracted from the answer sentences of each document Based on the answer representative sentence, a second storage unit that stores an answer representative sentence in which a document corresponding to a plurality of source answer sentences is associated with one answer representative sentence, and each document associated with the one inquiry representative sentence Counts the number of documents that match each document associated with each of the answer representative sentences, and using the counted number of documents, FAQ candidates for a pair of one inquiry representative sentence and one answer representative sentence A FAQ candidate control unit for generating price information, a FAQ creation control unit for generating a FAQ composed of an inquiry and its answer based on a pair of an inquiry representative sentence and an answer representative sentence corresponding to the FAQ candidate evaluation information, It has a, FAQ candidate control unit is a matrix diagram of arranging the query representative sentence and answer representative sentence vertically and horizontally to a position intersecting the query representative sentence and the answer representative sentence aspect, query representative sentence applicable and answers A FAQ candidate matrix diagram in which FAQ candidate evaluation information corresponding to a pair with the representative sentence is displayed is generated, and the FAQ creation control unit selects the FAQ candidate evaluation information on the FAQ candidate matrix diagram displayed on a predetermined computer. Each of the source query sentences associated with the query representative sentence corresponding to the FAQ candidate evaluation information selected from the document set of the document including the query sentence and the answer sentence. Extracts and transmits the FAQ creation screen generates a computer including a query representative sentence and answer representative sentence corresponding to FAQ candidate evaluation information selected and extracted and the query statement of the extraction source.

It is a figure which shows the example of application of the FAQ preparation assistance system of 1st Embodiment. It is a figure explaining the FAQ preparation assistance function of 1st Embodiment. It is a block diagram of the FAQ creation support system of the first embodiment. It is a figure which shows an example of the inquiry-correspondence content 2-axis matrix figure of 1st Embodiment. It is a figure which shows the example of FAQ preparation screen of 1st Embodiment. It is a figure which shows the FAQ information example of 1st Embodiment. It is a figure which shows the FAQ creation assistance process flow of 1st Embodiment. It is a figure which shows the block block diagram (a) of the representative sentence production | generation part of 1st Embodiment, and the representative sentence production | generation process flow (b). It is a figure explaining the parsing (dependency structure tree) process which produces | generates the typical sentence of the inquiry content and reply content of 1st Embodiment. It is a figure which shows the example of extraction of the partial dependence structure tree in the syntax analysis process of 1st Embodiment. It is a figure which shows an example of the representative sentence obtained through the syntax analysis process of 1st Embodiment. It is a figure which shows an example of the representative sentence candidate information of 1st Embodiment. It is a figure which shows an example of the aggregation representative sentence candidate information of 1st Embodiment. It is a figure which shows an example of the inquiry representative sentence information of 1st Embodiment. It is a figure which shows an example of the reply representative text information of 1st Embodiment. It is a figure which shows an example of the inquiry representative sentence hierarchy information of 1st Embodiment. It is a figure which shows the example of a case information input screen of 1st Embodiment. It is a figure which shows an example of case information of 1st Embodiment.

  Hereinafter, embodiments will be described with reference to the drawings.

(First embodiment)
FIG. 1 to FIG. 18 are diagrams related to the FAQ creation support system of the first embodiment. As an example, the FAQ creation support system according to the present embodiment can be applied to a contact center system externally or internally, and a FAQ creation support function based on the contents of an inquiry from a customer and the contents of an operator's response (response). I will provide a.

  FIG. 1 is an example in which the FAQ creation support system 3 is applied to the contact center system 1. In the contact center system 1, the ACD system 2 distributes incoming calls (incoming calls) from the customer C via the public line network N to each operator, and the operator responds to the customer's inquiry.

  The operator inputs the customer inquiry content (inquiry text) from the operator terminal 5 and also inputs the response content (answer text) to the inquiry. Each inquiry and answer from the customer is registered as case information in a predetermined database and stored as a response history.

  FIG. 17 is an example of a case information input screen displayed on the display device of the operator terminal 5, and the operator inputs various response history information including inquiry contents and response contents from the customer through the case information input screen.

  Case information input from the operator terminal 5 through the case information input screen is stored in a storage unit such as a predetermined database. FIG. 18 is an example of case information, and includes a case number (document ID) uniquely assigned to one inquiry, an inquiry type, inquiry contents, corresponding contents, incidental information, and the like. Case information is generated and registered for one inquiry from the same or different customers. Note that the incidental information is information incidental to each case information such as a product classification and an inquiry classification for classifying inquiry contents.

  The FAQ creation support system 3 according to the present embodiment analyzes case information queries and responses stored as operator response history, generates (extracts) each query statement and each representative sentence of the response content, and generates Provide a FAQ creation environment that quantitatively evaluates query and answer pairs that are FAQ candidates from representative sentences.

  The case information registration process via the operator terminal 5 can be performed by a control unit (not shown) of the contact center system 1 or can be configured to be performed by the FAQ creation support system 3.

  FIG. 2 is a diagram for explaining the FAQ creation support function of this embodiment. From the case information including the inquiry contents from the customer inputted by the operator from the operator terminal 5 and the answer contents, a representative sentence generation process is performed through a syntax analysis of the inquiry sentence for each case number (document ID). Is generated. In addition, for the correspondence content, a representative sentence generation process is performed through syntactic analysis of the correspondence contents for each document ID to generate a representative answer sentence.

  The query representative sentence of the present embodiment is a summary sentence generated from a plurality of query sentences in the document set of the document including the query sentence and its answer sentence. This is a representative summary sentence that is a candidate sentence of FAQ in which a plurality of query sentences (and document IDs of extraction sources) are associated with one summary sentence. The same is true for the answer representative.

  Further, in the present embodiment, the syntactic analysis is further performed on the document group of the generation source associated with the generated representative sentence, a sub representative sentence associated with one representative sentence is generated, and the representative sentence and its sub representative are generated. A hierarchical representative sentence composed of sentences can be generated. In other words, by executing the representative sentence generation process for a plurality of inquiry contents associated with one representative sentence, the sub representative sentence is generated as a lower hierarchy of the representative sentence.

  In the document set of the document including the query sentence and the answer sentence, a plurality of query representative sentences (query summary sentences) appropriately representing the first document group composed of the query sentence of each document, and the answer sentence of each document Matching a plurality of response representative sentences (answer summary sentences) that appropriately represent the second document group composed of, the relationship between the query representative sentence and the answer representative sentence is represented by the number of document IDs (number of documents) of the extraction source A query-answer matrix diagram is generated.

  Specifically, the query representative sentence is the vertical axis and the answer representative sentence is the horizontal axis, and the number of matching between the document ID associated with the query representative sentence and the document ID associated with the reply representative sentence (the document ID number, that is, the document 2) A matrix that displays a two-axis display screen, and a query-response two-axis matrix diagram in which a pair of a query representative sentence and a representative answer sentence is evaluated by the number of document IDs of the extraction source. Etc. to the administrator terminal 4.

  In addition, when each display block that displays the number of document IDs where the vertical and horizontal axes of the query-answer two-axis matrix diagram intersect is selected, the query representative sentence that intersects and the answer representative sentence are displayed. An FAQ creation screen including the same is provided to the administrator terminal 4 and a source document group (query text of each document) associated with a pair of the query representative sentence corresponding to the selected display block and the representative answer sentence is provided. Realize FAQ creation environment.

  The FAQ created through the FAQ creation screen is registered as FAQ information. This FAQ information can be used as feedback information for the operator's inquiry handling work. For example, when the “FAQ Search” button on the case information input screen of FIG. 17 is selected, the registered FAQ information is displayed on the operator terminal, and the inquiry content is displayed. In addition, it is possible to reduce the burden of the input work, and to speed up the answer to the inquiry and to reduce the burden of the answer content input work.

  FIG. 3 is a configuration block diagram of the FAQ creation support system 3. The FAQ creation support system 3 includes a FAQ creation support server 100 and a DB server 200. The FAQ creation support server 100 functions as a control unit that controls the entire FAQ creation support function, and the DB server 200 functions as a storage unit that stores various types of information used for FAQ creation. The FAQ creation support system of this embodiment can be configured by one or a plurality of computers.

  The FAQ creation support server 100 controls the communication control unit 120 that performs communication control with the administrator terminal 4 (the FAQ creator terminal operated by the FAQ creator), the memory 130, and the entire FAQ creation support server. CPU (control unit) 110.

  The control unit 110 includes an authentication unit 111, a representative sentence generation unit 112, an FAQ candidate control unit 113, and an FAQ creation control unit 114.

  The DB server 200 stores screen information 210, case information 220, representative sentence information 230, and FAQ information 240, respectively. The representative sentence information 230 is a query representative sentence 231 generated through representative sentence generation processing by the representative sentence generation unit 112. , A hierarchical inquiry representative sentence 232 and an answer representative sentence 233.

<Representative sentence generation processing>
The representative sentence generation processing will be described with reference to FIGS. The representative sentence generation processing of this embodiment generates a representative sentence (summary) of an inquiry and a representative sentence of an answer by using a document set of documents (case information) including the inquiry sentence and the answer sentence as input data.

  As shown in FIG. 8, the representative sentence generation unit 112 includes a syntax analysis unit 1121, a representative sentence candidate extraction unit 1122, a sentence generation aggregation unit 1123, and a representative sentence determination unit 1124, and an extraction rule storage unit 250. including. The extraction rule storage unit 250 may be configured to be included in the DB server 200 or may be included in the FAQ creation support system 3 as a separate storage unit.

  The representative sentence generation unit 112 refers to the case information 220 shown in FIG. 18, analyzes the inquiry sentence (text format) received from the customer input by the operator stored in the inquiry field of each case information as one document, and analyzes it. A first document set of a query in which query statements of all target sources are collected is generated. Similarly, a second document set of answers obtained by collecting all answers of the analysis source is generated. The original sentence ID is associated with the inquiry sentence and the answer sentence of each document set.

  In this embodiment, using a parsing process and an extraction rule, a query representative sentence is generated from the query document set, and a reply representative sentence is generated from the answer document set.

  The parsing process of this embodiment analyzes a plurality of independent words constituting a sentence and a dependency relationship between the independent words. In the analysis result, a plurality of independent words and dependency relationships between the independent words are expressed using nodes and arcs. The parsing result of this embodiment is generated as a dependency structure tree.

  A node represents an independent word in the dependency structure tree. This node is given a headword of the independent word, a part of speech of the headword, and an adjunct to the headword. The headword of the independent word given to the node indicates a character string of the independent word. The part of speech of the headword given to the node represents the part of speech of the headword (that is, the independent word represented by the node). The part of speech given to a node includes, for example, a noun, a sa variable noun, a verb, an adjective, an adverb and a conjunction.

  An adjunct word attached to a node represents an adjunct word accompanying the entry word. The adjunct to the headword given to the node includes particles such as “GA”, “NO”, “NO” and “NI”.

  An arc represents a syntactic dependency between nodes in a dependency structure tree. This arc is given a dependency type between nodes (independent words). The types of dependency relationships given to the arc include, for example, ga rating, wo rating, linkage modification, and adjacency. In the dependency structure tree, the arc is described by an arrow, for example. It is assumed that the arc arrow points from the source node to the destination node in the dependency relationship between the nodes.

  In the following description, in a dependency relationship between two nodes represented by using one arc, a node at the destination of the arc (that is, a node that is an end point in one arc) is referred to as a parent node. On the other hand, in a dependency relationship between two nodes represented by using one arc, a dependency source node in the arc (that is, a node that is a starting point in one arc) is referred to as a child node.

  FIG. 9A is an example of a dependency structure tree expressed using arcs representing two nodes and a dependency relationship between the nodes. In the dependency syntax tree of FIG. 9A, the node 201 and the node 202 are connected by the arc 203, the node 201 corresponds to the parent node, and the node 202 corresponds to the child node. By combining dependency structure trees as shown in FIG. 9A, a syntax analysis result (dependency structure tree) of a sentence including a plurality of independent words is expressed.

  FIG. 9B shows a syntax analysis of the first sentence “suddenly black ink does not come out and cannot be printed” of the query sentences constituting the document with the document ID “1” in FIG. It is an example of a result.

  The root node is a node that does not have a parent node. In the example of FIG. 9B, the root word is a node that cannot be entered. Further, a child node for the root node is defined as a first generation child node. In the example of FIG. 9B, the first generation child node has a headword of “no longer appearing” and a headword of “print”. Is a node. Then, a node having no child node (that is, not connected to the child node by an arc) is defined as a leaf node. In the example of FIG. 9B, the headwords are “suddenly”, “black”, “print”. Each node corresponds to a leaf node.

  In the present embodiment, each query statement in each document constituting the document set is individually parsed, and representative sentence candidates (representative sentence candidates) are extracted from the dependency structure tree obtained as an analysis result. . Rules for extracting representative sentence candidates are stored in advance in the extraction rule storage unit 250 as representative sentence candidate extraction rules. In the present embodiment, one or more representative sentence candidates are extracted from the query sentence by applying the representative sentence candidate extraction rule to the dependency structure tree obtained as an analysis result.

  The representative sentence candidate extraction process using the representative sentence candidate extraction rule will be described. In the present embodiment, by applying each of a plurality of extraction rules to one dependency structure tree, a partial dependency structure tree is extracted from the dependency structure tree for each extraction rule. Further, each of the extraction rules is applied to the extracted partial dependency structure tree, whereby the partial dependency structure tree is extracted. That is, in the representative sentence candidate extraction process, each partial structure tree extracted for each extraction rule becomes a representative sentence candidate.

  10A and 10B are diagrams for explaining representative sentence candidate extraction rules. FIG. 10A shows a rule for extracting a partial dependency structure tree including one verb (extraction rule 1), and FIG. 10B shows a branch. It is explanatory drawing of the rule (extraction rule 2) which extracts a none part dependence structure tree.

  The extraction rule 1 is a partial dependency structure tree rule including one verb, and as shown in FIG. 10A, focuses on a verb among a plurality of independent words represented by the dependency structure tree. According to the extraction rule 1, the dependency structure tree is divided based on a verb among a plurality of independent words represented by the dependency structure tree.

  Specifically, in the dependency structure tree to which the extraction rule 1 is applied, the dependency is obtained by cutting the arc between the node whose part of speech of the headword given to the node is a verb and the parent node of the verb node. The structural tree is split. That is, in the extraction rule 1, each divided dependency structure tree is extracted as a partial dependency structure tree.

  The extraction rule 2 is a rule for extracting a partial dependency structure tree without a branch, and as shown in FIG. 10B, the root node and the first generation child node (that is, the dependency tree) (that is, the partial dependency structure tree). Pay attention to all types of arcs between child nodes of the root node). In these arcs, the extraction rule 2 is applied when there are arcs of which the types of arc are ga, wo, ni, kara, place, and tool. The type of arc to which this extraction rule 2 is applied is set in advance.

  First, arcs whose arc types are ga, wo, ni, kara, place, and tool are searched from all arcs between the root node and the first generation child nodes in the partial dependency structure tree. Is done.

  Next, among all arcs between the root node and the first generation child node, arcs whose arc types are other than ga, wo, ni, kara, place, and tool (that is, arcs searched for) Other arcs) are cut.

  Thereafter, in the partial dependency structure tree after the arc is cut, a partial dependency structure tree including all nodes and arcs between the root node and each leaf node is extracted. In the example of FIG. 10B, three partial dependency structure trees are extracted.

  In addition, in the extraction rule 2, a partial dependency structure tree without a branch is extracted from the partial dependency structure tree, and the partial dependency structure tree extracted by applying the extraction rule 2 becomes a branch-less dependency structure tree.

  FIG. 11 shows a partial dependency extracted by applying the extraction rule 1 and / or the extraction rule 2 to the dependency structure tree of the sentence “suddenly black ink does not come out and cannot be printed” shown in FIG. 9B. It is an example of a structural tree. In this way, one or more partial dependency structure trees, that is, representative sentence candidates in the dependency structure tree format, are extracted from one query sentence using syntax analysis and extraction rules. These representative sentence candidates are associated with a representative sentence candidate ID, an extraction source document ID, and an extraction source field (inquiry, answer, etc.).

  As shown in FIG. 11, when the extraction rule 1 is applied to the parsing result (dependency structure tree) of FIG. 9B, the dependency structure tree is divided by two verbs “cannot be output” and “cannot be performed”. Two partial dependency structure trees A and B are extracted. Further, the extraction rule 1 and the extraction rule 2 are applied to the syntax analysis result (dependency structure tree) of FIG. When applied, the partial dependency structure tree C is extracted from the partial dependency structure tree A, and the partial dependency structure tree D is extracted from the partial dependency structure tree C.

  Then, each of the plurality of partial dependency structure trees A, B, C, and D is extracted as a representative sentence candidate of one query sentence, and a representative sentence candidate sentence is generated from the representative sentence candidate. That is, each representative sentence candidate sentence converted from the dependency structure tree form into the sentence form is generated based on the syntactic dependency relation (arc) between the nodes.

  For example, as shown in FIG. 11, a representative sentence candidate sentence can be generated from each partial dependency structure tree by setting the head word of the verb node as a final form. From the partial dependency structure tree B, the representative sentence candidate sentence “cannot print”, from the partial dependency structure tree C, the representative sentence candidate sentence “no black ink comes out”, from the partial dependency structure tree D A representative sentence candidate sentence “cannot be printed” is generated.

  FIG. 12 is a diagram illustrating an example of a generated representative sentence candidate sentence. A representative sentence candidate sentence is associated with a representative sentence candidate sentence ID, an extraction source document ID, and an extraction source field. Stored by candidate sentence ID.

  For example, the representative sentence candidate sentence of the representative sentence candidate sentence ID1 is “no black ink suddenly comes out”, the document ID “1” is associated, and the representative sentence candidate sentence ID2 is associated with the document ID “1”. The sentence candidate sentence is “cannot print”. Further, the representative sentence candidate sentence of the representative sentence candidate sentence ID4 associated with the document ID “1” is “cannot be printed”. In this way, a plurality of representative sentence candidate sentences are associated with one document ID, and a plurality of representative sentence candidates (representative sentence candidate sentences) generated from one document (query sentence) are managed by the document ID of the extraction source. .

  Next, a process of aggregating query sentences having the same character string constituting the representative sentence candidate sentence is performed to generate an aggregated representative sentence candidate sentence. A set of document IDs (number of document IDs) associated with the aggregated representative sentence candidate sentence is a union of document ID groups associated with the aggregated representative sentence candidate sentence.

  FIG. 13 is a diagram illustrating an example of an aggregate representative sentence candidate sentence in which a plurality of representative sentence candidate sentences shown in FIG. 12 are aggregated, and the aggregate representative sentence candidate sentence “2” cannot be printed with the aggregate representative sentence candidate sentence ID “2”. The set of document IDs associated with "is" 1, 2, ... ".

  That is, the process of aggregating representative sentence candidate sentences is performed by searching for representative sentence candidate sentences associated with other document IDs using character strings constituting representative sentence candidate sentences associated with one document ID, and different document IDs. Is associated with one representative sentence candidate sentence.

  Then, the representative sentence is generated by extracting the representative sentence representing the representative contents of the extraction source document set using the aggregated representative sentence candidate sentence.

  For example, as an example of a representative content level (representativeness), the number of different document IDs, that is, the number of source document IDs (number of documents) associated with the representative sentence candidate sentence can be used and aggregated. As the number of document IDs associated with the representative sentence candidate sentence is larger, it can be extracted (determined) as a sentence representing the contents of the document set having a higher representative degree.

  Accordingly, a predetermined number (for example, 10) of aggregated representative sentence candidate sentences are extracted in descending order of the number of documents (representativeness) using the number of document IDs associated with each of the aggregated representative sentence candidate sentences of the query in FIG. Then, the predetermined number of extracted representative representative sentence candidate sentences can be determined as representative sentences. FIG. 14 is a diagram illustrating an example of a representative sentence (summary) of a query determined from the aggregated representative sentence candidate sentences. Each representative sentence is generated by document ID (representative sentence) for each representative sentence ID. Each document ID) associated with the determined aggregated representative sentence candidate sentence is associated.

  As described above, the representative sentence generation processing according to the present embodiment performs the query sentence and the answer sentence of the document set from the document set (a plurality of documents including the query sentence and the answer sentence) including the query sentence and the answer sentence. Each representative sentence that appropriately represents each is generated. Each representative sentence is associated with each document of the document set as a generation source, and each different document is associated with the representative sentence (different documents are aggregated (classified) by the representative sentence).

  FIG. 8B is a diagram showing a representative sentence generation processing flow of the present embodiment.

  The syntax analysis unit 1121 extracts each document including the query sentence of the case information 220 and the answer sentence as a pair by document ID (S301), and targets a document set including query sentences including a plurality of independent words. The parsing process of each query statement is performed (S302).

  The syntax analysis unit 1121 performs syntax analysis on each query sentence that constitutes each document included in the target document set, that is, each query sentence included in each document included in the document set. The result of the syntax analysis by the syntax analysis unit 1121 is expressed by a dependency structure tree. Note that the result of parsing one query statement is one dependency structure tree, and a dependency structure tree is generated for each query statement.

  The representative sentence candidate extraction unit 1122 extracts a representative sentence candidate that is a partial structure tree that is a part of the dependency structure tree generated by the syntax analysis unit 1122 from the dependency structure tree (step S303). The representative sentence candidate extraction unit 1122 extracts representative sentence candidates using the extraction rules stored in the extraction rule storage unit 250. The representative sentence candidate (partial dependency structure tree) extracted by the representative sentence extracting unit 1122 is a structure tree that represents at least two independent words and a dependency relationship between the independent words. The representative sentence candidate extraction unit 1122 extracts representative sentence candidates from each of the dependency structure trees generated for each query sentence by the syntax analysis unit 1121.

  The extraction rules stored in the extraction rule storage unit 250 are rules that can be applied to the dependency structure tree and extract representative sentence candidates from the dependency structure tree. In the present embodiment, a plurality of different extraction rules 1 and 2 are stored in the extraction rule storage unit 250, and the representative sentence candidate extraction unit 1122 of the present embodiment applies the extraction rule 2 after applying the extraction rule 1, and each part A representative sentence candidate that is a structure tree is extracted from the object-dependent structure tree.

  The sentence generation / aggregation unit 1123 determines the representative sentence candidate based on a plurality of independent words represented by the representative sentence candidate (partial structure tree) extracted by the representative sentence candidate extraction unit 1122 and the dependency relationship between the independent words. The representative sentence candidate sentence (plain text) is generated from (S304).

  Next, the sentence generation / aggregation unit 1123 generates an aggregated representative sentence candidate sentence by aggregating the generated representative sentence candidate sentences (step S305). The sentence generation / aggregation unit 1123 aggregates the same representative sentence candidate sentences into one aggregated representative sentence candidate sentence among the generated representative sentence candidate sentences. At this time, the sentence generation / aggregation unit 1123 sets each document ID associated with the representative sentence candidate sentence aggregated in the aggregated representative sentence candidate sentence for each aggregated representative sentence candidate sentence ID for identifying the aggregated representative sentence candidate sentence. Associate.

  The representative sentence determination unit 1124 determines (selects) a representative sentence from the aggregated representative sentence candidate sentences generated by the sentence generation / aggregation unit 1123 (step S306). At this time, the representative sentence determination unit 1124 determines the number of document IDs assigned to the aggregated representative sentence candidate sentences generated by the sentence generation / aggregation part 1123 (that is, the representative sentence candidate sentences aggregated in the aggregated representative sentence candidate sentences). The representative sentence is determined based on the number.

  The representative sentence determination unit 1124 assigns representative sentence identification IDs to a plurality of aggregate representative sentence candidate sentences determined as representative sentences, and assigns each document ID associated with the representative sentence candidate sentences aggregated in the aggregate representative sentence candidate sentences. The database is associated and stored in the DB server 200 (predetermined storage area) (S307).

  In addition, the representative sentence generation unit 112 extracts each document including the query sentence of the case information 220 and the answer sentence as a pair for each document ID, and targets a document group composed of answer sentences including a plurality of independent words. Performs representative sentence generation processing, generates a representative sentence (answer summary sentence) associated with the document ID of the generation source, and identifies a representative sentence to a plurality of aggregate representative sentence candidate sentences determined as the representative sentence In addition to assigning an ID, each document ID associated with the representative sentence candidate sentence aggregated with the aggregated representative sentence candidate sentence is associated and stored in the DB server 200 (predetermined storage area) (FIG. 15).

  In the present embodiment, a representative sentence generation process is further performed on a generation source document group associated with the generated query representative sentence, and a subclass query representative sentence (sub-query) for the generated one query representative sentence is processed. Representative sentence).

  For example, by performing the representative sentence generation processing for a plurality of documents with respective document IDs associated with the representative sentence “cannot be printed” assigned the representative sentence ID “1” in FIG. As shown, sub representative sentences such as “black ink cannot be produced” and “black cannot be produced” are generated for the representative sentence “cannot print”. In other words, a representative sentence generated for a group of documents containing the representative sentence “Cannot print” is associated as a sub representative sentence in the lower hierarchy of the representative sentence “Cannot print” in the upper hierarchy, and the query representative sentence. It is possible to generate a hierarchical query representative sentence including the hierarchical structure. The generated sub representative sentence is stored in the DB server 200 (predetermined storage area) in association with the upper representative sentence ID for each sub representative sentence ID.

<FAQ creation support function>
The FAQ creation support function of this embodiment will be described with reference to FIGS. The FAQ creation support function of the present embodiment uses the hierarchical query representative sentence and the answer representative sentence generated by the representative sentence generation process, and creates a FAQ through an inquiry-response matrix diagram (FAQ candidate matrix diagram) and a FAQ creation screen. The environment is provided to the administrator terminal 4 (FAQ creator).

  The FAQ creation control unit 114 performs FAQ creation processing based on the FAQ creation request transmitted from the administrator terminal 4 that has undergone authentication processing by the authentication unit 111.

  When the FAQ creation control unit 114 receives the FAQ creation request, the FAQ creation control unit 114 outputs, to the FAQ candidate control unit 113, an instruction to generate a query-answer matrix diagram using the hierarchical query representative sentence and the answer representative sentence.

  The FAQ candidate control unit 113 to which the generation instruction is input performs processing for generating FAQ candidate evaluation information and processing for generating an inquiry-answer matrix diagram. The FAQ candidate control unit 113 refers to the hierarchical query representative sentence 233 and the answer representative sentence 232 stored in the DB server 200, and associates each document related to one query representative sentence with each of the answer representative sentences. The number of documents matching each existing document is counted, and FAQ candidate evaluation information for a pair of one inquiry representative sentence and one answer representative sentence is generated using each counted document number.

The FAQ candidate evaluation information of this embodiment is the number of counted documents. For example, the document ID associated with the inquiry representative sentence “cannot be printed” is “1, 2, 5, 8, 9, 10, 11,. .., And the document ID associated with the reply representative sentence “Perform automatic error correction of printing” is “8, 11, 25, 31,...”, Two document IDs “8” And “11” are matched, and the number of document IDs is counted as “2”.
The

  Further, the FAQ candidate control unit 113 counts the number of documents in which each document associated with one sub-query representative sentence matches each document associated with each answer representative sentence, and uses the counted number of documents. FAQ candidate evaluation information for a pair of a sub inquiry representative sentence and an answer representative sentence is also generated.

  The FAQ candidate evaluation information (number of document IDs) generated by the FAQ candidate evaluation information generation process of the present embodiment is information indicating the relationship between one inquiry representative sentence and one answer representative sentence, and is an FAQ candidate control unit. 113 displays the generated FAQ candidate evaluation information (the number of document IDs counted) in a display block at a position where the row and column of the corresponding query representative sentence and the reply representative sentence intersect, and the query representative sentence and the reply representative sentence It is provided as evaluation information that quantitatively evaluates the pair. When the count number is “0”, “0” is displayed on the display block.

  The FAQ candidate control unit 113 generates FAQ candidate evaluation information between all answer representative sentences for each query representative sentence and sub-query representative sentence, and the query representative sentence and the answer representative sentence are arranged vertically and horizontally. A query-response matrix diagram is generated in which FAQ candidate evaluation information corresponding to a pair of the query representative sentence and the answer representative sentence is displayed at a position where the query representative sentence and the answer representative sentence intersect vertically and horizontally. . At this time, the FAQ candidate control unit 113 arranges the sub-query representative sentence of the query representative sentence together with the corresponding query representative sentence in the matrix diagram, and the sub-query at the position where the sub-query representative sentence and the answer representative sentence intersect vertically and horizontally. An inquiry-answer matrix diagram is generated so that FAQ candidate evaluation information corresponding to a pair of representative sentence and answer representative sentence is displayed.

  For example, the inquiry representative sentences in FIG. 16 are arranged by representative sentence ID and sub representative sentence ID on the vertical axis, and similarly, the answer representative sentences in FIG. 15 are arranged by answer representative sentence ID on the horizontal axis. In the example of FIG. 16, the sub-representative sentences in the lower hierarchy linked to the representative sentence ID “1” are arranged between the representative sentence ID “1” and the representative sentence ID “2”.

  FIG. 4 is an example of a two-dimensional query-answer matrix diagram displayed on the administrator terminal 4. Each query representative sentence (including sub-query representative sentences) is shown on the vertical axis (column), and each horizontal axis (row) is shown on the horizontal axis (row). Each answer representative sentence is arranged, and each grid-like display block that quantitatively represents the relationship between the query representative sentence and the answer representative sentence in terms of the number of document IDs is included.

  The FAQ creator of the manager terminal 4 uses an operation input means such as a mouse to display a grid-like number in which the number of document IDs representing the relationship between the query representative sentence and the reply representative sentence on the query-answer matrix diagram is displayed. Each display block can be selected. When each of the grid-like display blocks is selected by the FAQ creator, the FAQ creation control unit 114 provides the administrator terminal 4 with a FAQ creation screen based on the inquiry representative sentence and the answer representative sentence corresponding to the selected display block. To do.

  If the grid-like display block (FAQ candidate evaluation information) on the inquiry-answer matrix diagram is selected, the FAQ creation control unit 114 determines whether the number of document IDs of the selected display block is “1” or more. Is determined.

  When it is determined that the number of document IDs of the selected display block is “1” or more, the FAQ creation control unit 114 displays an FAQ creation screen including an inquiry representative sentence and an answer representative sentence corresponding to the selected display block. It is generated and transmitted to the administrator terminal 4. When it is determined that the number of document IDs of the selected display block is not “1” or more, that is, the number of document IDs of the selected display block is “0”, the FAQ creation control unit 114 passes through the FAQ creation screen. The FAQ preparation process based on the inquiry representative sentence and the answer representative sentence is not performed.

  At this time, the FAQ creation control unit 114 refers to the case information 220 and displays the query representative sentence corresponding to the display block (FAQ candidate evaluation information) selected from the document set of the document including the query sentence and the answer sentence. Each query source of each extraction source to be associated is extracted based on the document ID, and includes a query representative sentence and an answer representative sentence corresponding to the selected selected display block, and a query sentence of each extracted source. Generate a FAQ creation screen.

  FIG. 5 is an example of the FAQ creation screen of the present embodiment. In the example of FIG. 5, the display block “25” corresponding to the query representative sentence “cannot print” and the reply representative sentence “check ink remaining amount” in the query-answer matrix diagram of FIG. 4 is selected. It is a FAQ creation screen. Twenty-five query sentences (inquiry sentences associated with 25 document IDs) showing a quantitative evaluation (relationship) between the inquiry representative sentence “cannot print” and the answer representative sentence “check remaining ink level” Query list block A for displaying, query information of one document (document linked to document ID) selected in query list block A and the detailed information display block B for displaying the response text, query representative sentence "cannot print" Each of the reply representative sentences “confirm ink remaining amount” includes a new FAQ creation block C displayed in each input field.

  The FAQ creator sees a plurality of 25 source documents in the query list block that are categorized as a query representative sentence “cannot print” and a reply representative sentence “check remaining ink” on the FAQ creation screen. It is possible to view the inquiry sentence and the answer sentence of the one source document selected in the inquiry list block in the detailed information display block. The FAQ creation control unit 114 performs each control of display on the query list block A and display of the query text of one document selected in the query list block A and its answer text on the detailed information display block B.

  In addition, the FAQ creator edits the query representative sentence “Cannot print” displayed in the inquiry content input field of the new FAQ creation block, or the answer representative sentence “Check ink remaining amount” displayed in the corresponding content input field. By selecting the registration button, the FAQ creator makes a pair of the query text and the answer text displayed (input) in each of the query content input field and the corresponding content input field. FAQ can be registered (created).

  FIG. 6 is a diagram illustrating an example of FAQ information registered from the FAQ creation screen according to the present embodiment, and the FAQ creation control unit 114 uses a FAQ identification ID for each created FAQ based on a selection button selection operation. The subject name, the inquiry content, the inquiry representative sentence ID, the correspondence contents, and the answer representative sentence ID are stored as FAQ information 240. In the present embodiment, an FAQ creation environment based on the relationship between the query representative sentence and the reply representative sentence on the query-answer matrix diagram is provided, so that the FAQ creation screen is associated with the query representative sentence and the reply representative sentence. Become. For this reason, the FAQ information created through the FAQ creation screen includes the corresponding query representative sentence ID and answer representative sentence ID on the query-answer matrix diagram, respectively, and is stored in the FAQ information 240.

  The FAQ creation control unit 114 provides an FAQ creation screen including an inquiry representative sentence and an answer representative sentence corresponding to the selected display block to be provided to the administrator terminal 4 based on a selection operation to make the inquiry list block A. , Each extraction source query sentence associated with the query representative sentence corresponding to the display block (FAQ candidate evaluation information) selected from the document set of the document including the query sentence and the answer sentence with reference to the case information 220 Can be extracted based on the document ID, and the corresponding query sentence extracted can be displayed in the query list block A.

  Further, in the example of FIG. 4, it corresponds to the sub-representative representative sentence “document is put on hold” of the inquiry representative sentence “cannot print” in the inquiry-answer matrix diagram and the reply representative sentence “check remaining ink level”. When the display block is selected, the FAQ creation control unit 114 combines the selected sub-query representative sentence and the query representative sentence of the higher layer in the inquiry content input column of the new FAQ creation block C in FIG. "Cannot print. Document will be on hold" can be automatically generated and displayed as a FAQ candidate. That is, when a subclass query representative sentence is selected, a FAQ creation candidate (query content) is generated by combining the query representative sentence of the upper layer and the query representative sentence of the lower layer, and is automatically sent to the new FAQ creation block C. Can be displayed.

  FIG. 7 is a diagram illustrating a FAQ creation support process flow of the FAQ creation support server 100 according to the present embodiment.

  When receiving the FAQ creation request from the administrator terminal 4 (S201), the authentication unit 111 performs an authentication process (S101).

  Based on the FAQ creation request that has undergone the authentication process, the FAQ candidate control unit 113 acquires the hierarchical query representative sentence 233 and the answer representative sentence 232 stored in the DB server 200 (S102). , FAQ candidate evaluation information generation processing for generating FAQ candidate evaluation information between all answer representative sentences for each of a plurality of query representative sentences and / or sub-query representative sentences (S103), and a query representative sentence and a reply representative sentence A query-response matrix diagram is generated in which FAQ candidate evaluation information corresponding to the pair of the corresponding query representative sentence and the answer representative sentence is displayed at a position where the above and the other intersect each other vertically and horizontally (S104).

  The FAQ candidate control unit 113 transmits the generated inquiry-answer matrix diagram to the administrator terminal 4 (S105).

  When each grid-like display block displaying the number of document IDs representing the relationship between the inquiry representative sentence and the answer representative sentence on the inquiry-answer matrix diagram displayed on the administrator terminal 4 is selected (S106, S202). ), The FAQ creation control unit 114 extracts each source query sentence associated with the query representative sentence corresponding to the display block selected with reference to the case information 220 based on the document ID (S107). A FAQ creation screen including the inquiry representative sentence and the answer representative sentence corresponding to the displayed display block and the extracted inquiry sentence of each extraction source is generated (S108).

  The FAQ creation control unit 114 transmits the generated FAQ creation screen to the administrator terminal 4 and performs display control of the FAQ creation screen based on an operation input by the FAQ creator (S109). When the registration button is selected (S203), the FAQ creation control unit 114 sets a FAQ that is a pair of the query text and the answer text displayed (input) in each of the query content input field and the corresponding content input field. The information 240 is registered (S110).

  The FAQ creation support system according to the present embodiment includes a plurality of query representative sentences (query summary sentences) that appropriately represent the first document group composed of the query sentences of each document in a document set of documents including a query sentence and its answer sentence. ) And multiple answer representative sentences (answer summary sentences) that appropriately represent the second document group composed of the answer sentences of each document, and the relationship between the query representative sentence and the answer representative sentence is extracted By representing quantitatively by the number of document IDs, it is possible to provide an FAQ creation environment in which it is possible to easily and appropriately determine whether or not an FAQ is necessary.

  For example, the FAQ creator can quantitatively know how many events of the same content are registered regarding the inquiry and the answer, and can easily and appropriately determine whether or not the FAQ needs to be created.

  For this reason, it is possible to reduce the work time and work load compared to the work of conventionally reading a large number of response histories manually and generating FAQ candidates from scratch (from an empty state). In particular, since the FAQ candidates are expressed in sentences, it is possible to reduce the work of finishing the FAQ compared to the case of keywords, for example, the trouble of generating (creating) a document that is a FAQ candidate from the keywords, and the efficiency of FAQ creation Can be improved.

  Further, in the present embodiment, the query-answer matrix including each of the query representative sentences and the answer representative sentences arranged vertically and horizontally, and each grid-like display block indicating the quantitative relationship between the query representative sentences and the answer representative sentences. Since the figure is provided to the FAQ creator, it is possible to realize a quantitative evaluation in which the query sentence for creating the FAQ and the document including the answer sentence are appropriately classified, and the document including the query sentence and the answer sentence. It is possible to easily grasp at a glance the overall image of FAQ candidate sentences created from a set of documents (for example, the tendency of many queries and their responses).

  In addition, since the FAQ creation screen is provided based on the selection of the display block on the query-answer matrix diagram, the FAQ creation is performed on the document set including the query sentence and the answer sentence aggregated (classified) by quantitative evaluation. An environment can be provided. That is, since the FAQ creation screen itself is a FAQ creation environment that is aggregated into a document set by an inquiry representative sentence and an answer representative sentence, an appropriate FAQ can be created without departing from the subject matter of the FAQ.

  In addition, since each query source of each source associated with the query representative sentence is extracted and displayed along with the query representative sentence and answer representative sentence on the FAQ creation screen, the detailed nuances and background of the summarized document (representative sentence) can be displayed. FAQ can be created while grasping.

  As described above, in the above-described embodiment, an example in which the processing for generating the FAQ candidate evaluation information and the processing for generating the query-response matrix diagram based on the hierarchical query representative sentence 233 and the answer representative sentence 232 has been described. Based on the sentence 231 and the answer representative sentence 232, the FAQ candidate evaluation information generation process and the inquiry-answer matrix diagram generation process can be performed.

  In addition to the matrix diagram composed of a plurality of query representative sentences and a plurality of reply representative sentences shown in FIG. 4, the query-answer matrix diagram is composed of, for example, one query representative sentence and a plurality of reply representative sentences. It is also possible to generate a matrix diagram including only a pair of a query representative sentence and a reply representative sentence in which the matrix diagram and FAQ candidate evaluation information are equal to or greater than a predetermined value (the number of document IDs is equal to or greater than a predetermined number).

  In addition, although an example of two-stage subclassing that generates a subclass query representative sentence by performing a representative sentence generation process on a generation source document group associated with the query representative sentence, the present invention is not limited to this. It is also possible to perform a representative sentence generation process on the query representative sentences of the subclass to generate multi-stage subclassified query representative sentences of three levels, four levels,...

  In the above-described embodiment, the subclassification of the inquiry representative sentence has been described as an example. However, for example, the answer representative sentence can be subclassified. That is, it is possible to further generate a representative sentence generation process for the source document group associated with each of the answer representative sentences shown in FIG. 15, and generate a sub-answer representative sentence of a subclass corresponding to each answer representative sentence. In addition, it is possible to provide an FAQ creation environment in which either or both of the inquiry representative sentence and the answer representative sentence are subclassified. Note that, in the hierarchical matrix display of the answer representative sentences, the answer representative sentences are arranged by representative sentence ID and sub representative sentence ID on the horizontal axis in the example of FIG. The sub-representative sentences in the lower hierarchy associated with the answer representative sentence ID can be arranged between adjacent representative sentence IDs.

  In addition, subclassification of answer representative sentences can be performed in multiple stages of two or more layers. Further, similar to the processing when the sub-query representative sentence that is the lower hierarchy of the query representative sentence is selected in the query-answer matrix diagram, the upper-level answer representative sentence and the sub-answer representative sentence of the lower hierarchy in FIG. Can be generated and automatically displayed on the new FAQ creation block C.

  In addition, a representative for a document set of documents including a query sentence related to a specific product or a query type and an answer sentence using a product classification or a query classification included in each incidental information for each document ID of the case information 220. It can be configured to perform sentence generation processing, FAQ candidate evaluation information generation processing, or query-response matrix diagram generation processing, and provide an FAQ creation environment individually according to a specific product or inquiry type You can also.

  The representative sentence generation unit 112 of the FAQ creation support server 100 can be configured as an individual representative sentence generation apparatus, and is configured as an FAQ creation support system in which the FAQ creation support server 100 and the representative sentence generation apparatus are linked. You can also.

  Each process of the above-described embodiment can also be realized as a computer-executable program, and the computer in which the program is installed is information for performing each process of the FAQ creation support function according to the embodiment. It is possible to operate as a processing device. For example, the program is stored in an auxiliary storage device (not shown), and a control unit such as a CPU reads the program stored in the auxiliary storage device to the main storage device, and the control unit reads the program read to the main storage device. It is possible to execute and cause the computer to operate each process according to the embodiment.

  The program can be applied to a computer in a state where the program is recorded on a computer-readable recording medium, or can be downloaded to a computer through a network such as the Internet. Computer-readable recording media include optical disks such as CD-ROM, phase change optical disks such as DVD-ROM, magneto-optical disks such as MO (Magnet Optical) and MD (Mini Disk), floppy (registered trademark) disks, Examples include magnetic disks such as removable hard disks, memory cards such as compact flash (registered trademark), smart media, SD memory cards, and memory sticks. A hardware device such as an integrated circuit (IC chip or the like) specially designed and configured is also included as a recording medium.

  In addition, although some embodiment of this invention was described, these embodiment is shown as an example and is not intending limiting the range of invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

1 Contact Center System 2 ACD System 3 FAQ Creation Support System 4 Administrator Terminal 5 Operator Terminal 100 FAQ Creation Support Server 110 CPU (Control Unit)
111 Authentication Unit 112 Representative Sentence Generation Unit 1121 Syntax Analysis Unit 1122 Representative Sentence Candidate Extraction Unit 1123 Sentence Generation Aggregation Unit 1124 Representative Sentence Determination Unit 113 FAQ Candidate Control Unit 114 FAQ Creation Control Unit 120 Communication Control Unit 130 Memory 200 DB Server 210 Screen Information 220 Case information 230 Representative sentence information 231 Inquiry representative sentence information 232 Hierarchical inquiry representative sentence information 233 Answer representative sentence information 240 FAQ information 250 Extraction rule storage unit

Claims (3)

Based on the same query representative sentence among a plurality of query representative sentences extracted from the query sentences of each document in the document set of the document including the query sentence and the answer sentence, a plurality of extraction sources are included in one query representative sentence. A first storage unit for storing the query representative sentence associated with the document corresponding to the query sentence;
The answer in which the documents corresponding to the answer sentences of a plurality of extraction sources are associated with one answer representative sentence based on the same answer representative sentence among the answer representative sentences extracted from the answer sentences of the respective documents. A second storage unit for storing a representative sentence;
Each document associated with one query representative sentence counts the number of documents that match each document associated with each answer representative sentence, and one query representative sentence and one answer are counted using the counted number of each document. FAQ candidate control unit for generating FAQ candidate evaluation information for a pair with a representative sentence;
A FAQ creation control unit that generates a FAQ composed of a query and its answer based on a pair of a query representative sentence and a reply representative sentence corresponding to the FAQ candidate evaluation information. It is a matrix diagram in which the query representative sentence and the answer representative sentence are arranged vertically and horizontally, and at the position where the query representative sentence and the answer representative sentence intersect vertically and horizontally, the corresponding query representative sentence and answer representative sentence are paired. Generate a FAQ candidate matrix diagram displaying corresponding FAQ candidate evaluation information,
When the FAQ candidate evaluation information on the FAQ candidate matrix diagram displayed on a predetermined computer is selected, the FAQ creation control unit selects from the document set of documents including the inquiry sentence and the answer sentence. Each of the extraction source query sentences associated with the query representative sentence corresponding to the FAQ candidate evaluation information is extracted, and the query representative sentence and answer representative sentence corresponding to the selected FAQ candidate evaluation information are extracted. A FAQ creation support system that generates a FAQ creation screen including a query sentence of each extraction source and transmits it to the computer. The first storage unit includes one subquery based on the same subquery representative sentence among the subquery representative sentences for the query representative sentence extracted from the query sentences of a plurality of documents associated with the query representative sentence. Further storing the subquery representative sentence in which a document corresponding to the query sentence of a plurality of extraction sources is associated with the representative sentence;
The FAQ candidate control unit counts the number of documents in which each document associated with one subquery representative sentence matches each document associated with each answer representative sentence, and uses the counted number of documents to determine one. In addition to generating FAQ candidate evaluation information for a pair of a subquery representative sentence and one answer representative sentence,
The query representative sentence and the sub-query representative sentence of the query representative sentence and the answer representative sentence are arranged vertically and horizontally, the position where the query representative sentence and the answer representative sentence intersect vertically and horizontally, and the sub-query representative sentence FAQ candidate evaluation information corresponding to a pair of corresponding query representative sentence and answer representative sentence and FAQ candidate evaluation corresponding to a pair of sub-query representative sentence and answer representative sentence at each position where the answer representative sentence intersects vertically and horizontally The FAQ creation support system according to claim 1 , wherein the FAQ candidate matrix diagram displaying information is generated. Based on the same query representative sentence among a plurality of query representative sentences extracted from the query sentences of each document in the document set of the document including the query sentence and the answer sentence, a plurality of extraction sources are included in one query representative sentence. Based on the same answer representative sentence among a plurality of answer representative sentences extracted from the answer sentence of each of the documents, the first storage unit that stores the query representative sentence associated with the document corresponding to the query sentence, A computer connectable to a second storage unit that stores the answer representative sentences in which documents corresponding to a plurality of answer sentences of a plurality of extraction sources are associated with one answer representative sentence;
Each document associated with one query representative sentence counts the number of documents that match each document associated with each answer representative sentence, and one query representative sentence and one answer are counted using the counted number of each document. Based on a function for generating FAQ candidate evaluation information for a pair with a representative sentence and a pair of an inquiry representative sentence and an answer representative sentence corresponding to the FAQ candidate evaluation information, an FAQ including an inquiry and an answer is generated. And realize the function,
The function for generating the FAQ candidate evaluation information is a matrix diagram in which the inquiry representative sentence and the answer representative sentence are arranged vertically and horizontally, and at the position where the inquiry representative sentence and the answer representative sentence intersect vertically and horizontally, Generate a FAQ candidate matrix diagram displaying FAQ candidate evaluation information corresponding to a pair of inquiry representative sentence and answer representative sentence,
When the FAQ candidate evaluation information on the FAQ candidate matrix diagram displayed on a predetermined computer is selected, the function for generating the FAQ is selected from a document set of documents including the inquiry sentence and the answer sentence. Each of the extraction source query sentences associated with the query representative sentence corresponding to the selected FAQ candidate evaluation information is extracted, and the query representative sentence and answer representative sentence corresponding to the selected FAQ candidate evaluation information, and the extraction A FAQ creation support program that generates a FAQ creation screen including each extracted query message and transmits it to the computer.

Images