WO2019205705A1

WO2019205705A1 - Semantic-framework-based human-machine conversation method and system

Info

Publication number: WO2019205705A1
Application number: PCT/CN2018/124937
Authority: WO
Inventors: 蔡振华; 肖龙源; 谭玉坤; 李稀敏; 刘晓葳
Original assignee: 厦门快商通信息技术有限公司
Priority date: 2018-04-28
Filing date: 2018-12-28
Publication date: 2019-10-31
Also published as: CN108932278B; CN108932278A

Abstract

A semantic-framework-based human-machine conversation method and system. The method comprises: creating a topic forest structure tree according to a raw corpus and generating a semantic framework model using the topic forest structure tree, and mapping an entity attribute of the topic forest structure tree to a corresponding semantic slot in a semantic framework model; during a human-machine conversation, matching a visitor question in terms of a topic type, and filling the visitor question into a semantic slot in a semantic framework model corresponding to the topic type; then mapping the visitor question filled into the semantic slot in the semantic framework model to the entity attribute of the topic forest structure tree; and finally, the topic forest structure tree matching a question in a knowledge base according to the visitor question, and feeding back an answer corresponding to the matching question to a visitor. Therefore, an accurate and complete visitor question can be acquired, so that the accuracy of an answer is ensured and the communication efficiency is improved based on same, and active interaction with a visitor can be achieved, thereby improving the user experience.

Description

Man-machine dialogue method and system based on semantic framework

Technical field

The invention relates to the field of artificial intelligence technology, in particular to a human-machine dialogue method based on a semantic framework and a system using the same.

Background technique

With the popularization of Internet and e-commerce, and the development of artificial intelligence technology, intelligent customer service is becoming more and more common. Intelligent customer service is an industry-oriented application developed on the basis of large-scale knowledge processing. It involves large-scale knowledge processing technology, natural language understanding technology, knowledge management technology, automatic question answering system, reasoning technology, etc., and has industry versatility. It not only provides enterprises with fine-grained knowledge management technology, but also establishes a fast and effective technical means based on natural language for communication between enterprises and mass users. At the same time, it can also provide statistical analysis information required for refined management. Can greatly reduce the labor costs of the company in customer service.

The working principle of intelligent customer service is mainly based on the application of big data knowledge processing technology, that is, by extracting the keyword of the visitor to judge the problem of the visitor, and then matching the corresponding answer to the visitor from the knowledge base. The premise of getting an accurate answer is to be able to extract accurate and complete questions. However, due to Chinese language problems, there are often multiple expressions and multiple use habits in the same problem, resulting in unsatisfactory answers and problem words, or the inability to identify users, resulting in a decline in user experience.

Summary of the invention

In order to solve the above problems, the present invention provides a human-machine dialogue method and system based on a semantic framework, which analyzes the human-machine dialogue content through the mapping relationship between the theme forest structure tree and the semantic framework model, thereby ensuring accurate acquisition. Complete visitor questions to ensure the accuracy of the answers and improve communication efficiency.

In order to achieve the above object, the technical solution adopted by the present invention is:

A man-machine dialogue method based on a semantic framework, which comprises the following steps:

a. creating a theme forest structure tree according to the original corpus, and extracting entity attributes corresponding to each topic type in the theme forest structure tree;

b. generating a semantic framework model by using the theme forest structure tree, and mapping an entity attribute of the theme forest structure tree to a corresponding semantic slot in the semantic framework model;

c. When the human-machine dialogue, the topic type is matched to the visitor question, and the visitor question is filled into the semantic slot in the semantic framework model corresponding to the topic type;

d. mapping the guest problem of the semantic slot of the filled semantic framework model to the entity attribute of the topic forest tree;

e. The theme forest tree performs question matching from the knowledge base according to the visitor question, and feeds the answer corresponding to the matched question to the visitor.

Preferably, in the step d, it is further determined, according to the mapped topic forest tree, whether the visitor problem satisfies a preset condition; when the visitor problem satisfies a preset condition, the theme forest tree according to the The guest problem is matched from the knowledge base; when the visitor problem does not satisfy the preset condition, the theme forest tree feeds the judgment result to the dialogue robot at the front end.

Further, the entity attribute includes a necessary attribute and an optional attribute, the semantic slot includes a necessary semantic slot and an optional semantic slot; the preset condition is whether the necessary attribute is complete; and when the human-machine dialogue is performed, the subject of the visitor question is Matching of types and populating guest questions into the necessary semantic slots and/or optional semantic slots in the semantic framework model corresponding to the topic type; and the necessary semantic slots and optional semantic slots of the populated semantic framework model The visitor problem is mapped to the necessary attributes and optional attributes of the topic forest tree; and further determining whether the necessary attribute is complete according to the mapped topic forest tree; when the necessary attribute of the visitor problem is complete, The theme forest tree performs problem matching from the knowledge base according to the visitor problem; when the necessary attribute of the visitor problem is incomplete, the theme forest tree feeds back the necessary attributes missing to the front-end dialogue robot, The dialogue robot asks the visitor according to the missing necessary attributes to obtain all necessary genus of the subject type .

Preferably, the step a further includes:

A1. Collect the original corpus and perform subject clustering on the original corpus to obtain different types of topics;

A2. Identify and extract an entity relationship for each topic type, and determine an entity attribute of each topic type according to the entity relationship;

A3. Create a topic tree for each type of topic based on the entity attributes, and create a topic forest knowledge base for all topic types.

Further, in the step a1, the subject clustering is performed on the original corpus, and the theme extraction and topic classification are performed by using the LDA topic model tool.

Further, in the step a2, the identification and extraction of the entity relationship for each topic type is performed by parsing and semantically parsing the original corpus, and extracting the relationship between the entity information and the tagged entity information according to the parsing result.

Further, in the step a3, the topic structure tree includes current topic information and inter-topic association information, and all types of topics are indexed according to the inter-topic association information to obtain a topic forest knowledge base.

Preferably, in the step c, the word segmentation processing and keyword extraction are performed on the visitor problem, and the type of the topic to which the keyword belongs is matched according to the extracted keyword, and the guest question is filled into the semantic corresponding to the topic type. The necessary semantic slots and/or optional semantic slots in the framework model.

Further, in the step d, the necessary semantic slots of the filled semantic framework model and the guest problem of the optional semantic slot are mapped to the necessary attributes and optional attributes of the theme forest structure tree, and the extraction and extraction are performed. The keyword matches the necessary attribute and the optional attribute, and judges whether the necessary attribute is missing according to the matching result.

Correspondingly, the present invention also provides a human-machine dialogue system based on a semantic framework, which includes:

a topic tree creation module, which creates a topic forest tree according to the original corpus, and extracts an entity attribute corresponding to each topic type in the topic forest tree;

a semantic framework model generating module, which generates a semantic framework model by using the theme forest structure tree, and maps an entity attribute of the theme forest structure tree to a corresponding semantic slot in the semantic framework model;

a human-machine dialog module for matching a topic type to a guest question, and populating a guest question into a semantic slot in a semantic framework model corresponding to the topic type;

a problem matching module, configured to map a guest problem of a semantic slot of the filled semantic framework model to an entity attribute of the topic forest structure tree, where the topic forest tree performs problem matching from the knowledge base according to the guest problem ;

An answer feedback module for feeding back the answer corresponding to the matched question to the visitor.

The beneficial effects of the invention are:

(1) The present invention analyzes the content of human-machine dialogue through the mapping relationship between the theme forest tree and the semantic framework model, and can ensure accurate and complete visitor problems, so as to ensure the accuracy of the answer and improve communication on the basis of this. effectiveness.

(2) The present invention maps the necessary and optional attributes of the theme when creating the topic forest-based knowledge base, and maps with the necessary semantic slots and optional semantic slots in the semantic framework model to thereby invite visitors during human-machine dialogue The problem is to match the topic matching and necessary attributes and the questioning of the necessary attributes, so as to actively interact with the visitor and increase the user experience.

DRAWINGS

The drawings described herein are intended to provide a further understanding of the invention, and are intended to be a part of the invention. In the drawing:

1 is a schematic flow chart of a human-machine dialogue method based on a semantic framework according to the present invention;

2 is a schematic structural diagram of a human-machine dialogue system based on a semantic framework according to the present invention.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in FIG. 1 , a semantic framework-based human-machine dialogue method of the present invention includes the following steps:

Among them, the semantic framework is a kind of knowledge representation. Frame Semantics is a cognitive linguistic theory proposed by American linguist Fillmore. The slot is the "slot" in the framework. Framework semantics is first and foremost a way to understand and describe the meaning of words and grammatical structures. It starts with the assumption that in order to understand the meaning of words in the language, we must first have the conceptual structure, that is, the knowledge of the semantic framework. The semantic framework provides the context and motivation for the meaning of words in the language and in the words. Framework semantics assumes that words can select and highlight certain aspects or instances of the basic semantic framework through its linguistic structure, which is done in a certain way (according to certain principles). Therefore, the interpretation of the meaning and function of words can be carried out in the light of the description of the basic semantic framework until the characteristics of these methods are detailed.

The semantic framework of the Chinese dictionary is as follows:

[Word form]: go

[Pinyin]: zou3

[producer]: Chen Qunxiu

[Work Order Number]: 6

[verb type]: his verb

[Number of arguments]: 2

[Number of meanings]: 10**

[Sense item number]: 10*

[Interpretation]: * "going" in the meaning of metaphor.

[Basic 1]: Shi Shi + Go + Guest

[Basic sentence example 1]: China unswervingly follows the path of reform and opening up. The construction of the teaching team has embarked on the track of standardization and legalization.

[Basic 2]: Shi Shi + Go + Direction

[Basic sentence example 2]: The Chinese football team goes to the world. Domestic elevators are heading for the world. The three national epics went to the bookshelf. The sailing era is coming to an end.

[Name of the thesis]: Acting

[syntax function]: subject

[Semantic classification]: {super class|things|objects|spacetime}

[Thesis]: Universe, national economy, industry, Chinese football team, domestic elevator, sailing era, works

[Name of the thesis]: Guest

[syntax function]: object

[semantic classification]: {space|abstracts}

[Example of thesis]: road, track, orbit

[Name of the thesis]: direction

[syntax function]: adverbial

[semantic classification]: {space|abstract|specific}

[mark of the purpose]: toward; on;

[Thesis]: the world, bookshelves, end, light, implementation, the rule of law, failure, stability, traitorism

[Negative]: No V V can't

[Temperature]: To 1V, V will be V, V will be V V, V is going, V is V, V is going down, V is down, V is up 2

[Remarks 1]: Extended: Location

The brackets represent the semantic slot, and the content after the colon represents the content filled by the semantic slot. The semantic slot can be divided into the necessary semantic slot and the optional semantic slot as needed.

In this embodiment:

In the step a, the entity attribute includes a required attribute and an optional attribute.

In the step b, the semantic slot includes a necessary semantic slot and an optional semantic slot.

In the step c, during the human-machine dialogue, the topic type is matched to the guest question, and the guest question is filled into the necessary semantic slot and/or the optional semantic slot in the semantic framework model corresponding to the topic type; The necessary semantic slots of the populated semantic framework model and the guest questions of the optional semantic slot are mapped to the necessary and optional attributes of the topic forest tree.

In the step d, it is further determined, according to the mapped topic forest tree, whether the visitor problem satisfies a preset condition; when the visitor question satisfies a preset condition, the topic forest tree is based on the visitor problem The problem matching is performed from the knowledge base; when the guest question does not satisfy the preset condition, the theme forest tree feeds the judgment result to the dialogue robot at the front end. In this embodiment, the preset condition is whether the necessary attribute is complete; that is, determining whether the required attribute is complete according to the mapped topic forest tree; and when the necessary attribute of the visitor problem is complete, the theme forest The tree performs question matching from the knowledge base according to the visitor question; when the necessary attribute of the visitor question is incomplete, the theme forest tree feeds back the necessary attributes that are missing to the front-end dialog robot, and the dialog robot The visitor is questioned based on the missing necessary attributes to get all the necessary attributes of the subject type. In this embodiment, after the challenge, the process returns to step c to perform the extraction of the guest question, the filling of the semantic slot, the mapping of the attribute of the entity, the determination of the missing attribute, and the like, and repeats the above process until all the required trees of the theme forest tree are satisfied. Required attributes and retrieve the answers from the knowledge base to the visitor.

Wherein, the step a further includes:

details as follows:

In the step a1, subject clustering of the original corpus is to use the LDA topic model tool for topic extraction and topic classification. The original corpus refers to a historical conversation record between the visitor and the customer service, and the original corpus is periodically updated or updated in real time according to the new conversation record. The LDA (Latent Dirichlet Allocation) topic model is a document topic generation model, also called a three-layer Bayesian probability model, which includes a three-layer structure of words, topics and documents. The document to topic follows a polynomial distribution, and the subject to the word follows a polynomial distribution. For each document, a topic is extracted from the topic distribution, and a word is extracted from the word distribution corresponding to the extracted topic; the above process is repeated until each word in the document is traversed, thereby obtaining the subject of the document. The document is a dialogue record between the visitor and the customer service in the present invention. For example, a raw corpus is divided into topics such as weather queries, train inquiries, flight inquiries, and the like.

In the step a2, the identification and extraction of the entity relationship for each topic type is performed by parsing and semantically parsing the original corpus, and extracting the relationship between the entity information and the tagged entity information according to the parsing result, and the entity may be used. The diagram is represented. Entity relationship diagram: A shorthand E-R diagram refers to the basic structure of data summarized by three basic concepts of entity, relationship and attribute. The entity is a named entity, which includes a text entity with explicit semantic information such as a name (organization name, person name, place name, trade name), an expression (date, time), etc., used in the ER diagram. The rectangle indicates that the name of the entity is written in the rectangle; for example, the visitor is an entity. The attribute, an attribute possessed by the entity, an entity can be characterized by several attributes; it is represented by an ellipse in the ER diagram, and is connected with the corresponding entity by an undirected edge; for example, the name of the visitor , account number, gender, etc., are attributes. The relationship refers to a way in which data objects are connected to each other, including a one-to-one relationship, a one-to-many relationship, and a many-to-many relationship.

In the step a3, the topic structure tree includes current topic information and inter-topic association information, and all types of topics are indexed according to the inter-topic relationship information to obtain a topic forest knowledge base. A conversation may be limited to a single topic within a domain, or it may involve multiple topics in multiple domains. By matching the topics involved in the guest question in the conversation, when a single topic is involved, the topic type is searched for by the topic forest knowledge base, and the necessary and optional attributes of the topic type are obtained to confirm the integrity of the problem. When multiple topics are involved, matching multiple topic types through the topic forest knowledge base, and obtaining corresponding necessary attributes and optional attributes of multiple topic types to confirm the visitor problem in each topic Integrity in the type.

In the step c, the topic type matching is performed on the visitor problem by performing word segmentation processing and keyword extraction on the visitor problem, matching the type of the topic to which the keyword is matched according to the extracted keyword, and filling the visitor question into the location. The necessary semantic slots and/or optional semantic slots in the semantic framework model corresponding to the topic type. In the step c, filling the guest question into the semantic slot in the semantic framework model corresponding to the topic type, using the natural language framework parser to populate the corresponding content in the guest question into each of the semantic framework models. In the semantic slot.

In the step d, the necessary semantic slots of the filled semantic framework model and the guest problem of the optional semantic slot are mapped to the necessary attributes and optional attributes of the theme forest structure tree, and the extracted keywords are Matching the necessary attributes and optional attributes, and judging whether the necessary attributes are missing according to the matching result.

In this embodiment, the weather query topic is taken as an example for explanation:

First, build the theme forest tree:

Root: weather query;

Required attribute 1: time;

Required attribute 2: location;

Optional attribute 1: Weather type, such as rain, snow, haze, etc.;

Optional attribute 2: Other.

Then, the semantic framework model is generated according to the theme forest tree as follows:

Weather query framework

Slot 1 time:

Slot 2 location:

Slot 3 weather type:

Slot 4 other:

Then, get the visitor question: Will it rain in Beijing tomorrow?

The results obtained after the weather query semantic framework model are as follows:

Semantic Framework: Weather Query

Slot 1 time: Tomorrow (the system can be determined according to the current system date);

Slot 2 Location: Beijing

Slot 3 Weather Type: Raining

Slot 4 other: none

The semantic framework model maps the above acquired content to the theme forest structure tree. After the theme forest structure tree is judged, it finds that it has met the necessary attribute 1 and the necessary attribute 2, so the problem query is made into the knowledge base and the query result (answer ) Feedback to visitors.

As shown in FIG. 2, the present invention also provides a human-machine dialogue system based on a semantic framework, which includes:

It should be noted that each embodiment in the specification is described in a progressive manner, and each embodiment focuses on differences from other embodiments, and the same similar parts between the embodiments are referred to each other. can. For the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

Also, the term "comprises", "comprising", or any other variants thereof, is intended to encompass a non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements but also Other elements not explicitly listed, or elements that are inherent to such a process, method, item, or device. An element that is defined by the phrase "comprising a ..." does not exclude the presence of additional equivalent elements in the process, method, item, or device that comprises the element. In addition, those skilled in the art can understand that all or part of the steps of implementing the foregoing embodiments may be completed by hardware, or may be instructed by a program to perform related hardware.

The above description shows and describes the preferred embodiments of the present invention. It is to be understood that the invention is not to be construed as being limited to the details disclosed herein. And modifications can be made by the above teachings or related art or knowledge within the scope of the inventive concept. All changes and modifications made by those skilled in the art are intended to be within the scope of the appended claims.

Claims

A man-machine dialogue method based on a semantic framework, characterized in that it comprises the following steps:

a. creating a theme forest structure tree according to the original corpus, and extracting entity attributes corresponding to each topic type in the theme forest structure tree;

b. generating a semantic framework model by using the theme forest structure tree, and mapping an entity attribute of the theme forest structure tree to a corresponding semantic slot in the semantic framework model;

c. When the human-machine dialogue, the topic type is matched to the visitor question, and the visitor question is filled into the semantic slot in the semantic framework model corresponding to the topic type;

d. mapping the guest problem of the semantic slot of the filled semantic framework model to the entity attribute of the topic forest tree;

e. The theme forest tree performs question matching from the knowledge base according to the visitor question, and feeds the answer corresponding to the matched question to the visitor.
The semantic framework-based human-machine dialogue method according to claim 1, wherein in the step d, determining whether the visitor problem satisfies a preset condition according to the mapped theme forest structure tree; When the visitor question satisfies a preset condition, the theme forest tree performs question matching from the knowledge base according to the visitor question; when the visitor question does not satisfy the preset condition, the theme forest tree will judge The result is fed back to the dialogue robot at the front end.
The semantic framework-based human-machine dialog method according to claim 2, wherein the entity attribute comprises a necessary attribute and an optional attribute, the semantic slot includes a necessary semantic slot and an optional semantic slot; The preset condition is whether the necessary attribute is complete; when the man-machine conversation is performed, the topic type is matched to the guest question, and the guest question is filled into the necessary semantic slot and/or the optional semantic slot in the semantic framework model corresponding to the topic type. And mapping the necessary semantic slots of the filled semantic framework model and the guest problem of the optional semantic slot to the necessary attributes and optional attributes of the topic forest structure tree; further determining according to the mapped theme forest tree Whether the required attribute is complete; when the necessary attribute of the visitor question is complete, the subject forest tree performs question matching from the knowledge base according to the visitor question; when the necessary attribute of the visitor question is incomplete, The theme forest tree feeds back the necessary attributes of the missing to the front-end dialogue robot, which is based on the missing The necessary attributes are asked by the visitor to get all the necessary attributes of the subject type.
The human-machine dialogue method based on the semantic framework according to any one of claims 1 to 3, wherein the step a further comprises:

A1. Collect the original corpus and perform subject clustering on the original corpus to obtain different types of topics;

A2. Identify and extract an entity relationship for each topic type, and determine an entity attribute of each topic type according to the entity relationship;

A3. Create a topic tree for each type of topic based on the entity attributes, and create a topic forest knowledge base for all topic types.
The semantic framework-based human-machine dialogue method according to claim 4, wherein in the step a1, subject clustering of the original corpus is performed by using the LDA topic model tool for topic extraction and topic classification.
The human-machine dialogue method based on the semantic framework according to claim 4, wherein in the step a2, the entity relationship is identified and extracted for each topic type, and the original corpus is parsed by syntax analysis. And semantic parsing, extracting the relationship between the entity information and the annotated entity information according to the parsing result.
The semantic framework-based human-machine dialog method according to claim 4, wherein in the step a3, the topic structure tree includes current topic information and inter-topic association information, according to the inter-topic association Information correlates all types of topics to get a topical forest knowledge base.
The semantic framework-based human-machine dialogue method according to claim 3, wherein in the step c, the word segmentation processing and keyword extraction are performed on the visitor problem, and the extracted keywords are matched. The topic type, and the guest question is populated into the necessary semantic slots and/or optional semantic slots in the semantic framework model corresponding to the topic type.
The semantic framework-based human-machine dialogue method according to claim 8, wherein the step d is a guest problem by using a necessary semantic slot of the filled semantic framework model and an optional semantic slot. Mapping to the necessary attributes and optional attributes of the topic forest tree, and matching the extracted keywords with the necessary attributes and optional attributes, and determining whether the necessary attributes are missing according to the matching result.
A human-machine dialogue system based on a semantic framework, which is characterized in that it comprises:

a topic tree creation module, which creates a topic forest tree according to the original corpus, and extracts an entity attribute corresponding to each topic type in the topic forest tree;

a semantic framework model generating module, which generates a semantic framework model by using the theme forest structure tree, and maps an entity attribute of the theme forest structure tree to a corresponding semantic slot in the semantic framework model;

a human-machine dialog module for matching a topic type to a guest question, and populating a guest question into a semantic slot in a semantic framework model corresponding to the topic type;

a problem matching module, configured to map a guest problem of a semantic slot of the filled semantic framework model to an entity attribute of the topic forest structure tree, where the topic forest tree performs problem matching from the knowledge base according to the guest problem ;

An answer feedback module for feeding back the answer corresponding to the matched question to the visitor.