WO2021229773A1 - Inquiry subject aggregation device, inquiry subject aggregation method, and program - Google Patents

Inquiry subject aggregation device, inquiry subject aggregation method, and program Download PDF

Info

Publication number
WO2021229773A1
WO2021229773A1 PCT/JP2020/019352 JP2020019352W WO2021229773A1 WO 2021229773 A1 WO2021229773 A1 WO 2021229773A1 JP 2020019352 W JP2020019352 W JP 2020019352W WO 2021229773 A1 WO2021229773 A1 WO 2021229773A1
Authority
WO
WIPO (PCT)
Prior art keywords
inquiry
label
predicate
aggregation device
statement
Prior art date
Application number
PCT/JP2020/019352
Other languages
French (fr)
Japanese (ja)
Inventor
のぞみ 小林
邦子 齋藤
準二 富田
Original Assignee
日本電信電話株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電信電話株式会社 filed Critical 日本電信電話株式会社
Priority to PCT/JP2020/019352 priority Critical patent/WO2021229773A1/en
Publication of WO2021229773A1 publication Critical patent/WO2021229773A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/38—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/383—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities

Definitions

  • the present invention relates to a technique for aggregating customer inquiries (requests) in a text chat response log.
  • the contact center plays an important role as a point of contact between the customer and the company, but in recent years, not only the conventional telephone-based contact center system but also the system by text chat is being introduced.
  • Non-Patent Document 1 discloses, as a conventional method, a method of parsing an inquiry sentence to create a "representative sentence" and aggregating it in units of representative sentences.
  • Non-Patent Document 1 the inquiry sentence is parsed, the representative sentence candidate is extracted by pattern extraction, and the representative sentence is the one in which the number of documents from which the representative sentence candidate is extracted is large. Therefore, a sentence asking "whether or not there is a merit” that can be regarded as the same as the content of the inquiry (eg. "Is there a demerit?" "Is there a merit?") And a sentence asking "possible / impossible” (e) .G. "Is it possible to make a contract?" "Is it possible to make a contract?”) Is a separate representative sentence, and there is a problem that it cannot be aggregated. That is, the conventional technique has a problem that the accuracy of aggregating inquiry sentences is low.
  • the present invention has been made in view of the above points, and an object of the present invention is to provide a technique for aggregating inquiry sentences with high accuracy.
  • a message aggregation device for aggregating a set of inquiry sentences into one or more groups based on an inquiry label.
  • An inquiry type determination unit that determines the inquiry type of each inquiry statement using a determination model,
  • a query abstraction section that extracts predicate argument pairs and functional expression labels from each query statement,
  • a matter aggregation device is provided that includes an inquiry label assigning unit that assigns an inquiry label to each inquiry statement based on the inquiry type, the preceding descriptive word pair, and the functional expression label for each inquiry statement.
  • FIG. 1 is a block diagram of a requirement aggregation device 100 according to an embodiment of the present invention.
  • the message aggregation device 100 in the present embodiment includes a text analysis unit 110, an inquiry type determination unit 120, an inquiry type determination model storage unit 130, an inquiry abstraction unit 140, and an inquiry label assigning unit 150. It has a standard notation dictionary storage unit 160 and an antonym dictionary storage unit 170. Details of each part will be described later.
  • the input to the message aggregation device 100 is a set of inquiry statements, and the output is a set of inquiry statements to which a content label is attached to each inquiry statement. It is assumed that the set of inquiry sentences to be input in the present embodiment is the customer's inquiry sentence in the response log of the text chat in the contact center or the like, but this is an example, and the present invention is based on the present invention. It is also possible to apply it to targets other than the text chat response log.
  • Examples (1) to (8) of input to the message aggregation device 100 are as follows.
  • inquiry texts with the same content label can be regarded as the same inquiry content and aggregated.
  • the above-mentioned inquiry sentences (1) to (4) are grouped into one of the inquiry sentences having the same label of "true / false Q, cancellation fee, cost", and the above-mentioned (6) to (7).
  • Inquiries are grouped into one of the inquiries with the same label, "True / False Q, Merit, Yes".
  • S101 Text analysis
  • the text analysis unit 110 takes a set of inquiry sentences as an input, performs morphological analysis and dependency analysis for each inquiry sentence, and obtains morphological information such as notation, part of speech, imperfect form, and dependency information. .. Morphological analysis and dependency analysis can be performed using well-known techniques.
  • the inquiry type determination unit 120 takes a set of inquiry sentences analyzed by S101 as an input, and determines the inquiry type of each sentence by using the inquiry type determination model prepared in advance.
  • the inquiry type determination model is stored in the inquiry type determination model storage unit 130, and the inquiry type determination unit 120 reads out the inquiry type determination model from the inquiry type determination model storage unit 130 and uses it.
  • the inquiry type determination model storage unit 130 may be outside the message aggregation device 100.
  • the inquiry type determination model a model learned by a well-known machine learning method from data to which the correct inquiry type is manually given (for example, SupportVector Machines whose features are bag-of-words in morpheme notation) may be used, or manually. You may use the rule created in.
  • the query abstraction unit 140 takes a set of query sentences analyzed in S101 as an input, and outputs a predicate / term pair and a label of a functional expression from each sentence.
  • the predicate argument pairs and functional expression labels for each of the following input examples (1) to (5) are as follows. Note that the functional expression label is not attached if it does not match the predetermined rule.
  • the query abstraction unit 140 uses the dependency information to extract the last predicate and the case or the case related to the predicate as a predicate argument pair for each analyzed query sentence.
  • the ga-case or ha-case that appears at the end is the extraction target. If the predicate is a verb, it is extracted in the imperfect form. If the sa-variant noun is a predicate, it is extracted with the expression " ⁇ ".
  • the wo case is extracted. For example, in the case of an inquiry sentence "I want to start a contract in March", "contract, start” is a predicate argument pair.
  • the extracted information may be referred to as a "predicate term pair" by interpreting that "nothing as a term” is extracted as a term.
  • the query abstraction unit 140 assigns a functional expression label to a morpheme that appears after the predicate to be extracted as a predicate argument pair according to the following rule. If it matches multiple rules, multiple labels are combined with "_" and output. In some cases, the functional expression label is not attached. Even if the functional expression label is not attached, it may be interpreted that the functional expression label is attached in all cases by interpreting that "nothing as a functional expression label" is attached as the functional expression label. ..
  • the inquiry label assigning unit 150 includes a standard notation dictionary and antonym dictionary prepared in advance, a predicate argument pair and a functional expression label output by the inquiry abstraction unit 140, and an inquiry type output by the inquiry type determination unit 120. Based on this, an inquiry label is attached to each inquiry text.
  • the standard notation dictionary is a dictionary showing the standard notation for a certain notation, and is stored in the standard display dictionary storage unit 160.
  • the inquiry label assigning unit 150 uses the standard notation dictionary by referring to the standard display dictionary storage unit 160.
  • the standard display dictionary storage unit 160 may be provided outside the message aggregation device 100.
  • An example of a standard notation dictionary is shown in FIG.
  • the standard notation dictionary contains both expressions that correspond to "terms" and expressions that correspond to "predicates" in a predicate term pair.
  • the antonym dictionary is a dictionary showing the standard notation that is an antonym with respect to the standard notation of the "term" and the "predicate" of the predicate term pair, and is stored in the antonym dictionary storage unit 170.
  • the inquiry label assigning unit 150 uses the antonym dictionary by referring to the antonym dictionary storage unit 170.
  • An example of an antonym dictionary is shown in FIG.
  • the inquiry content (example: "is there a disadvantage” or “is there a merit”) that is the basis of each predicate argument pair of "standard notation” and "antonym” is Register the pair of "standard notation” and “antonym” in the antonym dictionary, which are related so that "standard notation” and “antonym” can be regarded as the same (eg, "whether or not there is merit”). It should be noted that such a registration method is an example.
  • the inquiry label assigning unit 150 first standardizes the predicate of the predicate term pair and the expression of each term according to the following flow.
  • the inquiry label assigning unit 150 determines the inquiry label by the following processing.
  • the predicate argument pair is "merit, yes”.
  • S3 Combines the information of the inquiry type, predicate argument pair, and functional expression label, and outputs it as an inquiry label.
  • FIG. 5 shows an example of the "inquiry type, predicate argument pair, functional expression label" input to the inquiry label assigning unit 150 and the inquiry label assigned as a result of processing.
  • the requirement aggregation device 100 in the present embodiment can be realized by, for example, causing a computer to execute a program in which the processing contents described in the present embodiment are described.
  • the "computer” may be a physical machine or a virtual machine on the cloud.
  • the "hardware” described here is virtual hardware.
  • the above program can be recorded on a computer-readable recording medium (portable memory, etc.), saved, and distributed. It is also possible to provide the above program through a network such as the Internet or e-mail.
  • FIG. 6 is a diagram showing an example of the hardware configuration of the above computer.
  • the computer of FIG. 6 has a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, which are connected to each other by a bus BS, respectively.
  • the program that realizes the processing on the computer is provided by, for example, a recording medium 1001 such as a CD-ROM or a memory card.
  • a recording medium 1001 such as a CD-ROM or a memory card.
  • the program is installed in the auxiliary storage device 1002 from the recording medium 1001 via the drive device 1000.
  • the program does not necessarily have to be installed from the recording medium 1001, and may be downloaded from another computer via the network.
  • the auxiliary storage device 1002 stores the installed program and also stores necessary files, data, and the like.
  • the memory device 1003 reads and stores the program from the auxiliary storage device 1002 when there is an instruction to start the program.
  • the CPU 1004 realizes the function related to the message aggregation device 100 according to the program stored in the memory device 1003.
  • the interface device 1005 is used as an interface for connecting to a network.
  • the display device 1006 displays a GUI (Graphical User Interface) or the like by a program.
  • the input device 1007 is composed of a keyboard, a mouse, buttons, a touch panel, and the like, and is used for inputting various operation instructions.
  • the output device 1008 outputs the calculation result.
  • the requirement aggregation device 100 makes it possible to aggregate inquiry sentences with higher accuracy than in the prior art.
  • a message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
  • An inquiry type determination unit that determines the inquiry type of each inquiry statement using a determination model,
  • a query abstraction section that extracts predicate argument pairs and functional expression labels from each query statement,
  • a matter aggregation device including an inquiry label assigning unit that assigns an inquiry label to each inquiry statement based on the inquiry type, the preceding descriptive word pair, and the functional expression label for each inquiry statement.
  • the query abstraction unit extracts the last predicate in the query sentence and the term related to the predicate as a predicate term pair, and extracts the functional expression label based on the expression of the morpheme appearing after the predicate.
  • the inquiry label assigning unit is the message aggregation device according to the first or second paragraph, in which each of the predicate and the term in the predicate term pair is replaced with the standard notation if there is an entry corresponding to the standard notation dictionary.
  • the inquiry label assigning unit is the corresponding standard when the standard notation of the predicate or the term in the predicate term pair of the inquiry sentence exists in the antonym dictionary.
  • the requirement aggregation device according to any one of the items 1 to 3 in which the notation is replaced with an antonym.
  • a message aggregation method executed by a message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
  • An inquiry type determination step that determines the inquiry type of each inquiry statement using the determination model, and A query abstraction step that extracts predicate argument pairs and functional expression labels from each query statement,
  • a matter aggregation method comprising a query label assignment step of assigning an inquiry label to each inquiry statement based on the inquiry type, the pre-descriptive word pair, and the functional expression label for each inquiry statement.
  • (Section 6) A program for making a computer function as each part in the requirement aggregation device according to any one of the items 1 to 4.
  • Subject aggregation device 110 Text analysis unit 120 Inquiry type judgment unit 130 Inquiry type judgment model storage unit 140 Inquiry abstraction unit 150 Inquiry label assignment unit 160 Standard notation dictionary storage unit 170 Anti-synonymous dictionary storage unit 1000 Drive device 1001 Recording medium 1002 Auxiliary Storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Machine Translation (AREA)

Abstract

An inquiry subject aggregation device for aggregating a set of inquiry sentences into one or more groups on the basis of an inquiry label, the inquiry content aggregation device comprising an inquiry type determination unit that determines the inquiry type of each inquiry sentence using a determination model, an inquiry abstraction unit that extracts a predicate-argument pair and a function expression label from each inquiry sentence, and an inquiry label application unit that applies an inquiry label to each inquiry sentence on the basis of the inquiry type, the predicate-argument pair, and the function expression label with respect to each inquiry sentence.

Description

用件集約装置、用件集約方法、及びプログラムSubject aggregation device, requirement aggregation method, and program
　本発明は、テキストチャット応対ログにおける顧客の問い合わせ内容（用件）を集約する技術に関連するものである。 The present invention relates to a technique for aggregating customer inquiries (requests) in a text chat response log.
　コンタクトセンタは顧客と企業の接点として重要な役割を担っているが、近年では、従来の電話ベースのコンタクトセンタシステムだけではなく、テキストチャットによるシステムの導入も進んでいる。 The contact center plays an important role as a point of contact between the customer and the company, but in recent years, not only the conventional telephone-based contact center system but also the system by text chat is being introduced.
　コンタクトセンタの業務改善においては顧客の「コンタクトリーズン（問い合わせをしてきた理由）」を把握することが重要であり、そのためには問い合わせ内容の抽出と同じ内容の問い合わせの集約が必要となる。特に、問い合わせ内容は表現のバリエーションが多く、同じとみなしたい問い合わせを集約することは重要な課題である。 In order to improve the business of the contact center, it is important to understand the customer's "contact reason (reason for making inquiries)", and for that purpose, it is necessary to collect inquiries with the same contents as the extraction of inquiries. In particular, there are many variations in the content of inquiries, and it is an important issue to aggregate inquiries that are considered to be the same.
　この課題に対し、非特許文献１に、従来の方法として、問い合わせ文を構文解析して「代表文」を作成し、代表文単位で集約する方法が開示されている。 For this problem, Non-Patent Document 1 discloses, as a conventional method, a method of parsing an inquiry sentence to create a "representative sentence" and aggregating it in units of representative sentences.
　非特許文献１に開示された従来技術では、問い合わせ文を構文解析し、パタン抽出によって代表文候補を抽出し、代表文候補が抽出された文書の数が多いものを代表文としている。そのため、問い合わせ内容として同じとみなせる「メリットの有無」を問うような文（ｅ．ｇ．「デメリットはありませんか」「メリットはありますか」）、「可能・不可能」を問うような文（ｅ．ｇ．「契約はできますか」「契約は可能ですか」）は別の代表文となり集約できないという課題がある。すなわち、従来技術では、問い合わせ文を集約する精度が低いという課題がある。 In the prior art disclosed in Non-Patent Document 1, the inquiry sentence is parsed, the representative sentence candidate is extracted by pattern extraction, and the representative sentence is the one in which the number of documents from which the representative sentence candidate is extracted is large. Therefore, a sentence asking "whether or not there is a merit" that can be regarded as the same as the content of the inquiry (eg. "Is there a demerit?" "Is there a merit?") And a sentence asking "possible / impossible" (e) .G. "Is it possible to make a contract?" "Is it possible to make a contract?") Is a separate representative sentence, and there is a problem that it cannot be aggregated. That is, the conventional technique has a problem that the accuracy of aggregating inquiry sentences is low.
　本発明は上記の点に鑑みてなされたものであり、問い合わせ文を高い精度で集約するための技術を提供することを目的とする。 The present invention has been made in view of the above points, and an object of the present invention is to provide a technique for aggregating inquiry sentences with high accuracy.
　開示の技術によれば、問い合わせ文の集合を問い合わせラベルに基づいて１以上のグループに集約するための用件集約装置であって、
　判定モデルを用いて各問い合わせ文の問い合わせタイプを判定する問い合わせタイプ判定部と、
　各問い合わせ文から述語項ペアと機能表現ラベルを抽出する問い合わせ抽象化部と、
　各問い合わせ文についての前記問い合わせタイプと前記述語項ペアと前記機能表現ラベルとに基づいて、各問い合わせ文に問い合わせラベルを付与する問い合わせラベル付与部と
　を備える用件集約装置が提供される。 According to the disclosed technology, it is a message aggregation device for aggregating a set of inquiry sentences into one or more groups based on an inquiry label.
An inquiry type determination unit that determines the inquiry type of each inquiry statement using a determination model,
A query abstraction section that extracts predicate argument pairs and functional expression labels from each query statement,
A matter aggregation device is provided that includes an inquiry label assigning unit that assigns an inquiry label to each inquiry statement based on the inquiry type, the preceding descriptive word pair, and the functional expression label for each inquiry statement.
　開示の技術によれば、応対ログの問い合わせ文を高い精度で集約することが可能となる。 According to the disclosed technology, it is possible to aggregate the inquiry texts of the response log with high accuracy.
本発明の実施の形態における用件集約装置の構成図である。It is a block diagram of the requirement aggregation apparatus in embodiment of this invention. 用件集約装置の動作を説明するためのフローチャートである。It is a flowchart for demonstrating operation of a matter aggregation apparatus. 標準表記辞書の例を示す図である。It is a figure which shows the example of the standard notation dictionary. 反義語辞書の例を示す図である。It is a figure which shows the example of the antonym dictionary. 問い合わせラベル付与部へ入力された問い合わせタイプ、述語項ペア、機能表現ラベルと、処理の結果付与される問い合わせラベルの例を示す図である。It is a figure which shows the example of the inquiry type, the predicate argument pair, the functional expression label which were input to the inquiry label assignment part, and the inquiry label which is attached as a result of processing. 用件集約装置のハードウェア構成例を示す図である。It is a figure which shows the hardware configuration example of the matter aggregation apparatus.
　以下、図面を参照して本発明の実施の形態（本実施の形態）を説明する。以下で説明する実施の形態は一例に過ぎず、本発明が適用される実施の形態は、以下の実施の形態に限られるわけではない。 Hereinafter, an embodiment of the present invention (the present embodiment) will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the following embodiments.
　（装置構成）
　図１は、本発明の実施の形態における用件集約装置１００の構成図である。図１に示すように、本実施の形態における用件集約装置１００は、テキスト解析部１１０、問い合わせタイプ判定部１２０、問い合わせタイプ判定モデル格納部１３０、問い合わせ抽象化部１４０、問い合わせラベル付与部１５０、標準表記辞書格納部１６０、反義語辞書格納部１７０を有する。各部の詳細については後述する。 (Device configuration)
FIG. 1 is a block diagram of a requirement aggregation device 100 according to an embodiment of the present invention. As shown in FIG. 1, the message aggregation device 100 in the present embodiment includes a text analysis unit 110, an inquiry type determination unit 120, an inquiry type determination model storage unit 130, an inquiry abstraction unit 140, and an inquiry label assigning unit 150. It has a standard notation dictionary storage unit 160 and an antonym dictionary storage unit 170. Details of each part will be described later.
　用件集約装置１００への入力は、問い合わせ文の集合であり、出力は、問い合わせ文毎に内容ラベルが付与された問い合わせ文の集合である。なお、本実施の形態において入力となる問い合わせ文の集合は、コンタクトセンタ等におけるテキストチャットの応対ログにおける顧客の問い合わせ文であることを想定しているが、これは例であり、本発明は、テキストチャットの応対ログ以外の対象に適用することも可能である。 The input to the message aggregation device 100 is a set of inquiry statements, and the output is a set of inquiry statements to which a content label is attached to each inquiry statement. It is assumed that the set of inquiry sentences to be input in the present embodiment is the customer's inquiry sentence in the response log of the text chat in the contact center or the like, but this is an example, and the present invention is based on the present invention. It is also possible to apply it to targets other than the text chat response log.
　用件集約装置１００への入力の例（１）～（８）は下記のとおりである。 Examples (1) to (8) of input to the message aggregation device 100 are as follows.
　（１）プラン変更を考えていますが、解約金はかかりますか？
　（２）プラン変更だけならば解約金はかからない？
　（３）プラン変更の場合に解約金は発生しますか？
　（４）プラン変更でも解約金は発生しない？
　（５）解約金はいつ支払うの？
　（６）デメリットはないのですか？
　（７）メリットはありますか？
　（８）追加で契約できますか？
　上記入力例に対応する、用件集約装置１００からの出力例（１）～（８）は下記のとおりである。下記の「真偽Ｑ、解約金、かかる」等が内容ラベルである。 (1) I am thinking of changing the plan, will there be a cancellation fee?
(2) Is there no cancellation fee if I just change the plan?
(3) Is there a cancellation fee if I change the plan?
(4) Will there be a cancellation fee even if I change the plan?
(5) When will you pay the cancellation fee?
(6) Are there any disadvantages?
(7) Is there any merit?
(8) Can I make an additional contract?
Output examples (1) to (8) from the message aggregation device 100 corresponding to the above input example are as follows. The following "true / false Q, cancellation fee, cost" etc. are the content labels.
　（１）真偽Ｑ、解約金、かかる：プラン変更を考えていますが、解約金はかかりますか？
　（２）真偽Ｑ、解約金、かかる：プラン変更だけならば解約金はかからない？
　（３）真偽Ｑ、解約金、かかる：プラン変更の場合に解約金は発生しますか？
　（４）真偽Ｑ、解約金、かかる：プラン変更でも解約金は発生しない？
　（５）日時Ｑ、解約金、支払う：解約金はいつ支払うの？
　（６）真偽Ｑ、メリット、ある：デメリットはないのですか？
　（７）真偽Ｑ、メリット、ある：メリットはありますか？
　（８）真偽Ｑ、契約、可能：追加で契約できますか？
　上記のように、同じ内容ラベルの問い合わせ文は同じ問い合わせ内容とみなして集約することができる。例えば、上記（１）～（４）の問い合わせ文は、「真偽Ｑ、解約金、かかる」という同じラベルを持つ問い合わせ文の１つにグループに集約され、上記（６）～（７）の問い合わせ文は、「真偽Ｑ、メリット、ある」という同じラベルを持つ問い合わせ文の１つにグループに集約される。 (1) Authenticity Q, cancellation fee, cost: I am thinking of changing the plan, will there be a cancellation fee?
(2) Authenticity Q, cancellation fee, cost: Is there no cancellation fee if only the plan is changed?
(3) Authenticity Q, cancellation fee, cost: Will there be a cancellation fee if the plan is changed?
(4) Authenticity Q, cancellation fee, cost: Does the cancellation fee occur even if the plan is changed?
(5) Date and time Q, cancellation fee, payment: When will the cancellation fee be paid?
(6) True / False Q, Advantages, Yes: Are there any disadvantages?
(7) Authenticity Q, merit, there is: Is there any merit?
(8) Authenticity Q, contract, possible: Can I make an additional contract?
As described above, inquiry texts with the same content label can be regarded as the same inquiry content and aggregated. For example, the above-mentioned inquiry sentences (1) to (4) are grouped into one of the inquiry sentences having the same label of "true / false Q, cancellation fee, cost", and the above-mentioned (6) to (7). Inquiries are grouped into one of the inquiries with the same label, "True / False Q, Merit, Yes".
　以下、図２のフローチャートの手順に沿って、用件集約装置１００の動作例を説明する。なお、Ｓ１０２とＳ１０３の処理の順番は逆でもよい。また、Ｓ１０２とＳ１０３を同時に実行してもよい。 Hereinafter, an operation example of the message aggregation device 100 will be described according to the procedure of the flowchart of FIG. The order of processing of S102 and S103 may be reversed. Further, S102 and S103 may be executed at the same time.
　（Ｓ１０１：テキスト解析）
　Ｓ１０１において、テキスト解析部１１０は、問い合わせ文の集合を入力とし、各問い合わせ文に対して、形態素解析と係り受け解析を実施し、表記、品詞、終止形等の形態素情報と係り受け情報を得る。形態素解析と係り受け解析については周知の技術を用いて行うことができる。 (S101: Text analysis)
In S101, the text analysis unit 110 takes a set of inquiry sentences as an input, performs morphological analysis and dependency analysis for each inquiry sentence, and obtains morphological information such as notation, part of speech, imperfect form, and dependency information. .. Morphological analysis and dependency analysis can be performed using well-known techniques.
　（Ｓ１０２：問い合わせタイプ判定）
　Ｓ１０２において、問い合わせタイプ判定部１２０は、Ｓ１０１により解析済みの問い合わせ文の集合を入力とし、予め用意した問い合わせタイプ判定モデルを用いて、各々の文の問い合わせタイプを判定する。図１の例では、問い合わせタイプ判定モデル格納部１３０に問い合わせタイプ判定モデルが格納されており、問い合わせタイプ判定部１２０は、問い合わせタイプ判定モデル格納部１３０から問い合わせタイプ判定モデルを読み出して使用する。なお、問い合わせタイプ判定モデル格納部１３０は、用件集約装置１００の外部にあってもよい。 (S102: Inquiry type determination)
In S102, the inquiry type determination unit 120 takes a set of inquiry sentences analyzed by S101 as an input, and determines the inquiry type of each sentence by using the inquiry type determination model prepared in advance. In the example of FIG. 1, the inquiry type determination model is stored in the inquiry type determination model storage unit 130, and the inquiry type determination unit 120 reads out the inquiry type determination model from the inquiry type determination model storage unit 130 and uses it. The inquiry type determination model storage unit 130 may be outside the message aggregation device 100.
　問い合わせタイプ判定モデルとして、人手で正しい問い合わせタイプを付与したデータから周知の機械学習手法で学習したモデル（例えば形態素表記のｂａｇ－ｏｆ－ｗｏｒｄｓを素性としたＳｕｐｐｏｒｔＶｅｃｔｏｒＭａｃｈｉｎｅｓ等）を用いてもよいし、人手で作成したルールを用いてもよい。 As the inquiry type determination model, a model learned by a well-known machine learning method from data to which the correct inquiry type is manually given (for example, SupportVector Machines whose features are bag-of-words in morpheme notation) may be used, or manually. You may use the rule created in.
　問い合わせタイプの定義として、例えば「５Ｗ１Ｈ」にあたる「場所Ｑ（どこ）」「日時Ｑ（いつ）」「人Ｑ（誰）」「モノＱ（何）」「方法Ｑ（どうやって）」「選択Ｑ（どちら）」、「真偽Ｑ（ＹＥＳ／ＮＯを答えとして想定する質問）」、「願望（～したい）」、「エラー報告（～できない）」、「依頼」からなるセットを利用できる。他に「推薦（おすすめは）」等を追加してもよい。 As the definition of the inquiry type, for example, "place Q (where)", "date and time Q (when)", "person Q (who)", "thing Q (what)", "method Q (how)", and "selection Q (selection Q)" corresponding to "5W1H". You can use a set consisting of "Which)", "True / False Q (question that assumes YES / NO as the answer)", "Wish (I want to)", "Error report (cannot)", and "Request". In addition, "recommendation (recommendation)" etc. may be added.
　（Ｓ１０３：問い合わせ抽象化）
　Ｓ１０３において、問い合わせ抽象化部１４０は、Ｓ１０１において解析済みの問い合わせ文の集合を入力とし、各々の文から述語と項のペアと機能表現のラベルを出力する。 (S103: Inquiry abstraction)
In S103, the query abstraction unit 140 takes a set of query sentences analyzed in S101 as an input, and outputs a predicate / term pair and a label of a functional expression from each sentence.
　下記の入力例（１）～（５）のそれぞれに対する述語項ペアと、機能表現のラベルは下記のとおりである。なお、機能表現のラベルについては、所定のルールにマッチしない場合には、付されない。 The predicate argument pairs and functional expression labels for each of the following input examples (1) to (5) are as follows. Note that the functional expression label is not attached if it does not match the predetermined rule.
　（１）プラン変更を考えていますが、解約金はかかりますか？
　述語項ペア：解約金、かかる
　（２）プラン変更だけならば解約金はかからない？
　述語項ペア：解約金、かかる
　機能表現ラベル：否定
　（３）プラン変更の場合に解約金は発生しますか？
　述語項ペア： 解約金、発生する
　（４）プラン変更でも解約金は発生しない？
　述語項ペア：解約金、発生する
　機能表現ラベル：否定
　（５）解約金はいつ支払うの？
　述語項ペア： 解約金、支払う　　
　以下、述語項ペア抽出と、機能表現ラベル抽出のそれぞれについて説明する。 (1) I am thinking of changing the plan, will there be a cancellation fee?
Predicate argument pair: Cancellation fee, it costs (2) Is there no cancellation fee if you just change the plan?
Predicate Argument Pair: Cancellation Fee, Functional Expression Label: Negation (3) Will there be a cancellation fee if the plan is changed?
Predicate Argument Pair: Cancellation fee will be incurred (4) Will cancellation fee be incurred even if the plan is changed?
Predicate argument pair: Cancellation fee, functional expression label generated: Negation (5) When will the cancellation fee be paid?
Predicate Argument Pair: Cancellation Fee, Pay
Hereinafter, each of the predicate argument pair extraction and the functional expression label extraction will be described.
　＜述語項ペア抽出＞
　問い合わせ抽象化部１４０は、各解析済みの問い合わせ文について、係り受け情報を用いて、最も末尾の述語と、その述語にかかるガ格もしくはハ格を述語項ペアとして抽出する。 <Predicate argument pair extraction>
The query abstraction unit 140 uses the dependency information to extract the last predicate and the case or the case related to the predicate as a predicate argument pair for each analyzed query sentence.
　対象の文において、該当するガ格もしくはハ格が複数存在した場合、最も末尾に出現したガ格もしくはハ格を抽出対象とする。なお、述語が動詞の場合は終止形で抽出する。また、サ変名詞が述語の場合、「○○する」という表現で抽出する。 In the target sentence, if there are multiple applicable ga-cases or ha-cases, the ga-case or ha-case that appears at the end is the extraction target. If the predicate is a verb, it is extracted in the imperfect form. If the sa-variant noun is a predicate, it is extracted with the expression "○○".
　ガ格もしくはハ格が存在せず、ヲ格が存在する場合はヲ格を抽出する。例えば、「３月に契約を開始したい」という問い合わせ文の場合、「契約、開始する」が述語項ペアとなる。 If there is no ga or ha case and there is a wo case, the wo case is extracted. For example, in the case of an inquiry sentence "I want to start a contract in March", "contract, start" is a predicate argument pair.
　ヲ格も存在しない場合は述語のみを抽出する。なお、述語のみを抽出する場合でも、"項として何もないこと"が項として抽出されたと解釈することで、抽出された情報を「述語項ペア」と呼ぶこととしてよい。 If the case does not exist, only the predicate is extracted. Even when only the predicate is extracted, the extracted information may be referred to as a "predicate term pair" by interpreting that "nothing as a term" is extracted as a term.
　項が存在せず、述語がサ変名詞の場合、「する」より前の表現を項として抽出する。例えば、「新規で契約できますか」の場合「契約する」が述語となり、「契約、する」を述語項ペアとして抽出する。 If the term does not exist and the predicate is a s-irregular noun, the expression before "do" is extracted as the term. For example, in the case of "Can I make a new contract?", "Contract" becomes a predicate, and "Contract, make" is extracted as a predicate argument pair.
　＜機能表現ラベル抽出＞
　問い合わせ抽象化部１４０は、述語項ペアとして抽出対象となった述語の後に出現する形態素に対し、下記ルールにより機能表現ラベルを付与する。複数のルールにマッチする場合は複数のラベルを「＿」で結合して出力する。なお、機能表現ラベルが付されない場合もある。機能表現ラベルが付されない場合でも、"機能表現ラベルとして何もないこと"が機能表現ラベルとして付されると解釈することで、全ての場合に機能表現ラベルが付されると解釈してもよい。 <Functional expression label extraction>
The query abstraction unit 140 assigns a functional expression label to a morpheme that appears after the predicate to be extracted as a predicate argument pair according to the following rule. If it matches multiple rules, multiple labels are combined with "_" and output. In some cases, the functional expression label is not attached. Even if the functional expression label is not attached, it may be interpreted that the functional expression label is attached in all cases by interpreting that "nothing as a functional expression label" is attached as the functional expression label. ..
　［ルールＡ］要望を表す表現（ｅ．ｇ．「たい」）が存在した場合、「要望」を付与する。 [Rule A] If there is an expression (eg "tai") that expresses a request, a "request" is given.
　［ルールＢ］否定を表す表現（ｅ．ｇ．「ない」「ません」「ぬ」「ず」）が存在した場合、「否定」を付与する。 [Rule B] If there is an expression indicating negation (eg "not", "not", "nu", "zu"), "denial" is given.
　［ルールＣ］可能を表す表現（ｅ．ｇ．「できる」）が存在した場合、「可能」を付与する。例えば、前述の例「新規で契約できますか」は「可能」が付与される。 [Rule C] If there is an expression (eg "can") that indicates possible, "possible" is given. For example, "possible" is given to the above example "Can I make a new contract?".
　（Ｓ１０４：問い合わせラベル付与）
　Ｓ１０４において、問い合わせラベル付与部１５０は、予め用意した標準表記辞書及び反義語辞書と、問い合わせ抽象化部１４０が出力する述語項ペア及び機能表現ラベルと、問い合わせタイプ判定部１２０が出力する問い合わせタイプとに基づいて、各々の問い合わせ文に問い合わせラベルを付与する。 (S104: Inquiry label assignment)
In S104, the inquiry label assigning unit 150 includes a standard notation dictionary and antonym dictionary prepared in advance, a predicate argument pair and a functional expression label output by the inquiry abstraction unit 140, and an inquiry type output by the inquiry type determination unit 120. Based on this, an inquiry label is attached to each inquiry text.
　標準表記辞書は、ある表記に対する標準表記を示す辞書であり、標準表示辞書格納部１６０に格納されている。問い合わせラベル付与部１５０は、標準表示辞書格納部１６０を参照することで標準表記辞書を利用する。なお、標準表示辞書格納部１６０は、用件集約装置１００の外部に備えられていてもよい。標準表記辞書の例を図３に示す。標準表記辞書には述語項ペアの「項」にあたる表現と「述語」にあたる表現の両方が含まれる。 The standard notation dictionary is a dictionary showing the standard notation for a certain notation, and is stored in the standard display dictionary storage unit 160. The inquiry label assigning unit 150 uses the standard notation dictionary by referring to the standard display dictionary storage unit 160. The standard display dictionary storage unit 160 may be provided outside the message aggregation device 100. An example of a standard notation dictionary is shown in FIG. The standard notation dictionary contains both expressions that correspond to "terms" and expressions that correspond to "predicates" in a predicate term pair.
　反義語辞書は、述語項ペアの「項」及び「述語」の標準表記に対し、反義語となる標準表記を示した辞書であり、反義語辞書格納部１７０に格納されている。問い合わせラベル付与部１５０は、反義語辞書格納部１７０を参照することで反義語辞書を利用する。反義語辞書の例を図４に示す。 The antonym dictionary is a dictionary showing the standard notation that is an antonym with respect to the standard notation of the "term" and the "predicate" of the predicate term pair, and is stored in the antonym dictionary storage unit 170. The inquiry label assigning unit 150 uses the antonym dictionary by referring to the antonym dictionary storage unit 170. An example of an antonym dictionary is shown in FIG.
　例えば、図４に示す例のように、「標準表記」と「反義語」のそれぞれの述語項ペアの元となる問い合わせ内容（例：「デメリットはないですか」「メリットはありますか」）が、「標準表記」と「反義語」とで同じ（例：「メリットの有無」を問う）と見なせるような関係にある「標準表記」と「反義語」の組を反義語辞書に登録する。なお、このような登録方法は一例である。 For example, as shown in the example shown in FIG. 4, the inquiry content (example: "is there a disadvantage" or "is there a merit") that is the basis of each predicate argument pair of "standard notation" and "antonym" is Register the pair of "standard notation" and "antonym" in the antonym dictionary, which are related so that "standard notation" and "antonym" can be regarded as the same (eg, "whether or not there is merit"). It should be noted that such a registration method is an example.
　問い合わせラベル付与部１５０は、まず述語項ペアの述語と項それぞれの表現の標準化を以下の流れで実施する。 The inquiry label assigning unit 150 first standardizes the predicate of the predicate term pair and the expression of each term according to the following flow.
　Ｓ１：述語項ペアの述語及び項のそれぞれについて標準表記辞書を引き、該当するエントリが存在すれば該当箇所を標準表記に置き換える。例えば、「解約金、発生する」は、「解約金、かかる」に置き換えられる。 S1: Look up the standard notation dictionary for each of the predicates and terms of the predicate argument pair, and if the corresponding entry exists, replace the relevant part with the standard notation. For example, "cancellation fee, incurred" is replaced with "cancellation fee, cost".
　Ｓ２：述語が「する」で機能表現ラベルに「可能」が存在する場合、もしくは、述語が「できる」の場合、述語を「可能」に置き換え、機能表現ラベルから「可能」を削除する。 S2: If the predicate is "do" and the functional expression label has "possible", or if the predicate is "possible", replace the predicate with "possible" and delete "possible" from the functional expression label.
　次に、問い合わせラベル付与部１５０は、下記の処理により問い合わせラベルを決定する。 Next, the inquiry label assigning unit 150 determines the inquiry label by the following processing.
　Ｓ１：問い合わせタイプが「真偽Ｑ」で、機能表現ラベルに「否定」を含む場合、機能表現ラベルから「否定」を削除する。 S1: If the inquiry type is "true / false Q" and the functional expression label contains "negative", "negative" is deleted from the functional expression label.
　例えば、ある問い合わせ文について、問い合わせタイプが「真偽Ｑ」であり、述語項ペアと機能表現ラベルが「デメリット、ある：否定」である場合、述語項ペアは「デメリット、ある」となる。 For example, for a certain inquiry sentence, if the inquiry type is "true / false Q" and the predicate argument pair and the functional expression label are "disadvantage, yes: negative", the predicate argument pair is "disadvantage, yes".
　Ｓ２：問い合わせタイプが「真偽Ｑ」で、述語もしくは項の標準表記が反義辞書に存在する場合、該当する標準表記を反義語に置き換える。 S2: If the inquiry type is "true / false Q" and the standard notation of the predicate or term exists in the antonym dictionary, the corresponding standard notation is replaced with the antonym.
　例えば、ある問い合わせ文について、問い合わせタイプが「真偽Ｑ」であり、述語項ペアが「デメリット、ある」である場合、述語項ペアは「メリット、ある」となる。 For example, for a certain inquiry sentence, if the inquiry type is "true / false Q" and the predicate argument pair is "disadvantage, yes", the predicate argument pair is "merit, yes".
　Ｓ３：問い合わせタイプ、述語項ペア、機能表現ラベルの情報を結合し、問い合わせラベルとして出力する。 S3: Combines the information of the inquiry type, predicate argument pair, and functional expression label, and outputs it as an inquiry label.
　図５に、問い合わせラベル付与部１５０へ入力された「問い合わせタイプ、述語項ペア、機能表現ラベル」と、処理の結果付与される問い合わせラベルの例を示す。 FIG. 5 shows an example of the "inquiry type, predicate argument pair, functional expression label" input to the inquiry label assigning unit 150 and the inquiry label assigned as a result of processing.
　（ハードウェア構成例）
　本実施の形態における用件集約装置１００は、例えば、コンピュータに、本実施の形態で説明する処理内容を記述したプログラムを実行させることにより実現可能である。なお、この「コンピュータ」は、物理マシンであってもよいし、クラウド上の仮想マシンであってもよい。仮想マシンを使用する場合、ここで説明する「ハードウェア」は仮想的なハードウェアである。 (Hardware configuration example)
The requirement aggregation device 100 in the present embodiment can be realized by, for example, causing a computer to execute a program in which the processing contents described in the present embodiment are described. The "computer" may be a physical machine or a virtual machine on the cloud. When using a virtual machine, the "hardware" described here is virtual hardware.
　上記プログラムは、コンピュータが読み取り可能な記録媒体（可搬メモリ等）に記録して、保存したり、配布したりすることが可能である。また、上記プログラムをインターネットや電子メール等、ネットワークを通して提供することも可能である。 The above program can be recorded on a computer-readable recording medium (portable memory, etc.), saved, and distributed. It is also possible to provide the above program through a network such as the Internet or e-mail.
　図６は、上記コンピュータのハードウェア構成例を示す図である。図６のコンピュータは、それぞれバスＢＳで相互に接続されているドライブ装置１０００、補助記憶装置１００２、メモリ装置１００３、ＣＰＵ１００４、インタフェース装置１００５、表示装置１００６、入力装置１００７、出力装置１００８等を有する。 FIG. 6 is a diagram showing an example of the hardware configuration of the above computer. The computer of FIG. 6 has a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, which are connected to each other by a bus BS, respectively.
　当該コンピュータでの処理を実現するプログラムは、例えば、ＣＤ－ＲＯＭ又はメモリカード等の記録媒体１００１によって提供される。プログラムを記憶した記録媒体１００１がドライブ装置１０００にセットされると、プログラムが記録媒体１００１からドライブ装置１０００を介して補助記憶装置１００２にインストールされる。但し、プログラムのインストールは必ずしも記録媒体１００１より行う必要はなく、ネットワークを介して他のコンピュータよりダウンロードするようにしてもよい。補助記憶装置１００２は、インストールされたプログラムを格納すると共に、必要なファイルやデータ等を格納する。 The program that realizes the processing on the computer is provided by, for example, a recording medium 1001 such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed in the auxiliary storage device 1002 from the recording medium 1001 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, and may be downloaded from another computer via the network. The auxiliary storage device 1002 stores the installed program and also stores necessary files, data, and the like.
　メモリ装置１００３は、プログラムの起動指示があった場合に、補助記憶装置１００２からプログラムを読み出して格納する。ＣＰＵ１００４は、メモリ装置１００３に格納されたプログラムに従って、用件集約装置１００に係る機能を実現する。インタフェース装置１００５は、ネットワークに接続するためのインタフェースとして用いられる。表示装置１００６はプログラムによるＧＵＩ（Ｇｒａｐｈｉｃａｌ　Ｕｓｅｒ　Ｉｎｔｅｒｆａｃｅ）等を表示する。入力装置１００７はキーボード及びマウス、ボタン、又はタッチパネル等で構成され、様々な操作指示を入力させるために用いられる。出力装置１００８は演算結果を出力する。 The memory device 1003 reads and stores the program from the auxiliary storage device 1002 when there is an instruction to start the program. The CPU 1004 realizes the function related to the message aggregation device 100 according to the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network. The display device 1006 displays a GUI (Graphical User Interface) or the like by a program. The input device 1007 is composed of a keyboard, a mouse, buttons, a touch panel, and the like, and is used for inputting various operation instructions. The output device 1008 outputs the calculation result.
　（実施の形態の効果）
　以上説明したとおり、本実施の形態に係る用件集約装置１００により、従来技術よりも高い精度で問い合わせ文を集約することが可能となる。 (Effect of embodiment)
As described above, the requirement aggregation device 100 according to the present embodiment makes it possible to aggregate inquiry sentences with higher accuracy than in the prior art.
　（実施の形態のまとめ）
　本明細書には、少なくとも下記の各項に記載した用件集約装置、用件集約方法、及びプログラムが記載されている。
（第１項）
　問い合わせ文の集合を問い合わせラベルに基づいて１以上のグループに集約するための用件集約装置であって、
　判定モデルを用いて各問い合わせ文の問い合わせタイプを判定する問い合わせタイプ判定部と、
　各問い合わせ文から述語項ペアと機能表現ラベルを抽出する問い合わせ抽象化部と、
　各問い合わせ文についての前記問い合わせタイプと前記述語項ペアと前記機能表現ラベルとに基づいて、各問い合わせ文に問い合わせラベルを付与する問い合わせラベル付与部と
　を備える用件集約装置。
（第２項）
　前記問い合わせ抽象化部は、問い合わせ文における最も末尾の述語と、当該述語に係る項とを述語項ペアとして抽出し、当該述語の後に出現する形態素の表現に基づいて機能表現ラベルを抽出する
　第１項に記載の用件集約装置。
（第３項）
　前記問い合わせラベル付与部は、述語項ペアにおける述語と項のそれぞれについて、標準表記辞書に該当するエントリがあれば標準表記に置き換える
　第１項又は第２項に記載の用件集約装置。
（第４項）
　問い合わせ文の問い合わせタイプが真偽を問うタイプである場合において、前記問い合わせラベル付与部は、当該問い合わせ文の述語項ペアにおける述語もしくは項の標準表記が反義辞書に存在する場合に、該当する標準表記を反義語に置き換える
　第１項ないし第３項のうちいずれか１項に記載の用件集約装置。
（第５項）
　問い合わせ文の集合を問い合わせラベルに基づいて１以上のグループに集約するための用件集約装置が実行する用件集約方法であって、
　判定モデルを用いて各問い合わせ文の問い合わせタイプを判定する問い合わせタイプ判定ステップと、
　各問い合わせ文から述語項ペアと機能表現ラベルを抽出する問い合わせ抽象化ステップと、
　各問い合わせ文についての前記問い合わせタイプと前記述語項ペアと前記機能表現ラベルとに基づいて、各問い合わせ文に問い合わせラベルを付与する問い合わせラベル付与ステップと
　を備える用件集約方法。
（第６項）
　コンピュータを、第１項ないし第４項のうちいずれか１項に記載の用件集約装置における各部として機能させるためのプログラム。 (Summary of embodiments)
This specification describes at least the message aggregation device, the message aggregation method, and the program described in each of the following sections.
(Section 1)
A message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
An inquiry type determination unit that determines the inquiry type of each inquiry statement using a determination model,
A query abstraction section that extracts predicate argument pairs and functional expression labels from each query statement,
A matter aggregation device including an inquiry label assigning unit that assigns an inquiry label to each inquiry statement based on the inquiry type, the preceding descriptive word pair, and the functional expression label for each inquiry statement.
(Section 2)
The query abstraction unit extracts the last predicate in the query sentence and the term related to the predicate as a predicate term pair, and extracts the functional expression label based on the expression of the morpheme appearing after the predicate. The requirements aggregation device described in the section.
(Section 3)
The inquiry label assigning unit is the message aggregation device according to the first or second paragraph, in which each of the predicate and the term in the predicate term pair is replaced with the standard notation if there is an entry corresponding to the standard notation dictionary.
(Section 4)
When the inquiry type of the inquiry sentence is a type that asks the truth, the inquiry label assigning unit is the corresponding standard when the standard notation of the predicate or the term in the predicate term pair of the inquiry sentence exists in the antonym dictionary. The requirement aggregation device according to any one of the items 1 to 3 in which the notation is replaced with an antonym.
(Section 5)
A message aggregation method executed by a message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
An inquiry type determination step that determines the inquiry type of each inquiry statement using the determination model, and
A query abstraction step that extracts predicate argument pairs and functional expression labels from each query statement,
A matter aggregation method comprising a query label assignment step of assigning an inquiry label to each inquiry statement based on the inquiry type, the pre-descriptive word pair, and the functional expression label for each inquiry statement.
(Section 6)
A program for making a computer function as each part in the requirement aggregation device according to any one of the items 1 to 4.
　以上、本実施の形態について説明したが、本発明はかかる特定の実施形態に限定されるものではなく、特許請求の範囲に記載された本発明の要旨の範囲内において、種々の変形・変更が可能である。 Although the present embodiment has been described above, the present invention is not limited to such a specific embodiment, and various modifications and changes can be made within the scope of the gist of the present invention described in the claims. It is possible.
１００　用件集約装置
１１０　テキスト解析部
１２０　問い合わせタイプ判定部
１３０　問い合わせタイプ判定モデル格納部
１４０　問い合わせ抽象化部
１５０　問い合わせラベル付与部
１６０　標準表記辞書格納部
１７０　反義語辞書格納部
１０００　ドライブ装置
１００１　記録媒体
１００２　補助記憶装置
１００３　メモリ装置
１００４　ＣＰＵ
１００５　インタフェース装置
１００６　表示装置
１００７　入力装置
１００８　出力装置 100 Subject aggregation device 110 Text analysis unit 120 Inquiry type judgment unit 130 Inquiry type judgment model storage unit 140 Inquiry abstraction unit 150 Inquiry label assignment unit 160 Standard notation dictionary storage unit 170 Anti-synonymous dictionary storage unit 1000 Drive device 1001 Recording medium 1002 Auxiliary Storage device 1003 Memory device 1004 CPU
1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims (6)

  1. 　問い合わせ文の集合を問い合わせラベルに基づいて１以上のグループに集約するための用件集約装置であって、
    　判定モデルを用いて各問い合わせ文の問い合わせタイプを判定する問い合わせタイプ判定部と、
    　各問い合わせ文から述語項ペアと機能表現ラベルを抽出する問い合わせ抽象化部と、
    　各問い合わせ文についての前記問い合わせタイプと前記述語項ペアと前記機能表現ラベルとに基づいて、各問い合わせ文に問い合わせラベルを付与する問い合わせラベル付与部と
    　を備える用件集約装置。     A message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
    An inquiry type determination unit that determines the inquiry type of each inquiry statement using a determination model,
    A query abstraction section that extracts predicate argument pairs and functional expression labels from each query statement,
    A matter aggregation device including an inquiry label assigning unit that assigns an inquiry label to each inquiry statement based on the inquiry type, the preceding descriptive word pair, and the functional expression label for each inquiry statement.
  2. 　前記問い合わせ抽象化部は、問い合わせ文における最も末尾の述語と、当該述語に係る項とを述語項ペアとして抽出し、当該述語の後に出現する形態素の表現に基づいて機能表現ラベルを抽出する
    　請求項１に記載の用件集約装置。     The query abstraction unit extracts the last predicate in the query sentence and the term related to the predicate as a predicate term pair, and extracts the functional expression label based on the expression of the morpheme appearing after the predicate. The requirement aggregation device according to 1.
  3. 　前記問い合わせラベル付与部は、述語項ペアにおける述語と項のそれぞれについて、標準表記辞書に該当するエントリがあれば標準表記に置き換える
    　請求項１又は２に記載の用件集約装置。     The requirement aggregation device according to claim 1 or 2, wherein the inquiry label assigning unit replaces each of the predicate and the argument in the predicate argument pair with the standard notation if there is an entry corresponding to the standard notation dictionary.
  4. 　問い合わせ文の問い合わせタイプが真偽を問うタイプである場合において、前記問い合わせラベル付与部は、当該問い合わせ文の述語項ペアにおける述語もしくは項の標準表記が反義辞書に存在する場合に、該当する標準表記を反義語に置き換える
    　請求項１ないし３のうちいずれか１項に記載の用件集約装置。     When the inquiry type of the inquiry sentence is a type that asks the truth, the inquiry label assigning unit is the corresponding standard when the standard notation of the predicate or the term in the predicate argument pair of the inquiry sentence exists in the antonym dictionary. The requirement aggregation device according to any one of claims 1 to 3, wherein the notation is replaced with an antonym.
  5. 　問い合わせ文の集合を問い合わせラベルに基づいて１以上のグループに集約するための用件集約装置が実行する用件集約方法であって、
    　判定モデルを用いて各問い合わせ文の問い合わせタイプを判定する問い合わせタイプ判定ステップと、
    　各問い合わせ文から述語項ペアと機能表現ラベルを抽出する問い合わせ抽象化ステップと、
    　各問い合わせ文についての前記問い合わせタイプと前記述語項ペアと前記機能表現ラベルとに基づいて、各問い合わせ文に問い合わせラベルを付与する問い合わせラベル付与ステップと
    　を備える用件集約方法。     A message aggregation method executed by a message aggregation device for aggregating a set of inquiry statements into one or more groups based on an inquiry label.
    An inquiry type determination step that determines the inquiry type of each inquiry statement using the determination model, and
    A query abstraction step that extracts predicate argument pairs and functional expression labels from each query statement,
    A matter aggregation method comprising a query label assignment step of assigning an inquiry label to each inquiry statement based on the inquiry type, the pre-descriptive word pair, and the functional expression label for each inquiry statement.
  6. 　コンピュータを、請求項１ないし４のうちいずれか１項に記載の用件集約装置における各部として機能させるためのプログラム。 A program for making a computer function as each part in the requirement aggregation device according to any one of claims 1 to 4.
PCT/JP2020/019352 2020-05-14 2020-05-14 Inquiry subject aggregation device, inquiry subject aggregation method, and program WO2021229773A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2020/019352 WO2021229773A1 (en) 2020-05-14 2020-05-14 Inquiry subject aggregation device, inquiry subject aggregation method, and program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2020/019352 WO2021229773A1 (en) 2020-05-14 2020-05-14 Inquiry subject aggregation device, inquiry subject aggregation method, and program

Publications (1)

Publication Number Publication Date
WO2021229773A1 true WO2021229773A1 (en) 2021-11-18

Family

ID=78525150

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/019352 WO2021229773A1 (en) 2020-05-14 2020-05-14 Inquiry subject aggregation device, inquiry subject aggregation method, and program

Country Status (1)

Country Link
WO (1) WO2021229773A1 (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2007087401A (en) * 2005-09-23 2007-04-05 Fuji Xerox Co Ltd System and method for indexing, and system and method and program for generating questionnaire template
US20120078891A1 (en) * 2010-09-28 2012-03-29 International Business Machines Corporation Providing answers to questions using multiple models to score candidate answers
JP2017091100A (en) * 2015-11-06 2017-05-25 日本電信電話株式会社 Predicate-argument structure extraction device, method, and program
JP2018206263A (en) * 2017-06-08 2018-12-27 日本電信電話株式会社 Predicate term structure model generation apparatus, predicate term structure analysis apparatus, method and program
JP2019175011A (en) * 2018-03-27 2019-10-10 日本電信電話株式会社 Response sentence generation apparatus, response sentence generation method, and program

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2007087401A (en) * 2005-09-23 2007-04-05 Fuji Xerox Co Ltd System and method for indexing, and system and method and program for generating questionnaire template
US20120078891A1 (en) * 2010-09-28 2012-03-29 International Business Machines Corporation Providing answers to questions using multiple models to score candidate answers
JP2017091100A (en) * 2015-11-06 2017-05-25 日本電信電話株式会社 Predicate-argument structure extraction device, method, and program
JP2018206263A (en) * 2017-06-08 2018-12-27 日本電信電話株式会社 Predicate term structure model generation apparatus, predicate term structure analysis apparatus, method and program
JP2019175011A (en) * 2018-03-27 2019-10-10 日本電信電話株式会社 Response sentence generation apparatus, response sentence generation method, and program

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
HIRANO, TOHRU ET AL.: "Personal Attribute Classification for Self- Disclosed Text for User Information Extraction", PROCEEDINGS OF THE 21ST ANNUAL MEETING OF THE NATURAL LANGUAGE PROCESSING SOCIETY; MARCH 16TH - 21ST, 2015, vol. 21, 16 February 2015 (2015-02-16), JP, pages 273 - 276, XP009532326 *

Similar Documents

Publication Publication Date Title
US10282468B2 (en) Document-based requirement identification and extraction
Leopold et al. Supporting process model validation through natural language generation
US5383120A (en) Method for tagging collocations in text
US10839134B2 (en) Attribution using semantic analysis
US8433560B2 (en) Rule based apparatus for modifying word annotations
US10877828B2 (en) Automatic analysis of a set of systems used to implement a process
JP6663826B2 (en) Computer and response generation method
CN110998590A (en) Domain-specific vocabulary-driven pre-parser
US20120078950A1 (en) Techniques for Extracting Unstructured Data
US11074402B1 (en) Linguistically consistent document annotation
Chatterji et al. AspectFrameNet: a frameNet extension for analysis of sentiments around product aspects
WO2021229773A1 (en) Inquiry subject aggregation device, inquiry subject aggregation method, and program
WO2021254227A1 (en) Targeted partial re-enrichment of a corpus based on nlp model enhancements
CN110998587A (en) Domain specific lexical analysis
JP2023540514A (en) Inference based on natural language interpretation
Adamu et al. A framework for enhancing the retrieval of UML diagrams
Dandapat et al. Training deployable general domain mt for a low resource language pair: English–bangla
JP2004287683A (en) Evaluation expression extraction device, program, storage medium, and evaluation expression extraction method
JP7064871B2 (en) Text mining device and text mining method
Al Ameri et al. Building lexical resources for dialectical Arabic
WO2021049485A1 (en) Legal analyzer and legal analysis method
JP3416918B2 (en) Automatic keyword extraction method and device
JP2011233163A (en) Sentence associating system and sentence associating program
US11354502B2 (en) Automated constraint extraction and testing
JP6651183B2 (en) Formal name / abbreviation list generation device, formal name / abbreviation list generation method, program

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20935273

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20935273

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP