CN114817490A

CN114817490A - Method and device for assisting foreign language intelligent conversation and intelligent conversation system

Info

Publication number: CN114817490A
Application number: CN202210009533.0A
Authority: CN
Inventors: 袁珊娜; 朴成杰; 谷海洋; 赵雪
Original assignee: Qingdao Haier Smart Technology R&D Co Ltd; Haier Smart Home Co Ltd
Current assignee: Qingdao Haier Smart Technology R&D Co Ltd; Haier Smart Home Co Ltd
Priority date: 2022-01-05
Filing date: 2022-01-05
Publication date: 2022-07-29

Abstract

The application relates to the technical field of artificial intelligence, and discloses a method for assisting foreign language intelligent conversation, which comprises the following steps: acquiring user dialogue information and extracting semantic key information; matching semantic associated information in a lower-layer common knowledge map according to the semantic key information; acquiring semantic expansion information from an upper-layer affair knowledge graph according to the semantic key information; and inputting the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information. The method and the device can realize association query and effective reasoning based on the user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience. The application also discloses a device for assisting the foreign language intelligent dialogue, an intelligent dialogue system and a storage medium.

Description

Method and device for assisting intelligent foreign language conversation and intelligent conversation system

Technical Field

The present application relates to the field of artificial intelligence technology, and for example, to a method and an apparatus for assisting intelligent foreign language dialogue, an intelligent dialogue system, and a storage medium.

Background

At present, with the improvement of education level, more and more people pay attention to foreign language teaching. To assist foreign language teaching, a number of smart question-answering systems and chatting robots have recently been introduced, which are capable of recognizing a user's question and generating a response to assist the user in learning a foreign language by matching answers to a corresponding knowledge base. However, such products have a high degree of dependence on the amount of data stored in the knowledge base, and often require a huge data set to be stored in advance to ensure the accuracy of the response, and the excessive data can seriously reduce the matching speed. Therefore, another technology is to construct a knowledge graph, wherein the knowledge graph comprises a first-level node, a second-level node, a third-level node and a fourth-level node; generating semantic slot rules for multi-turn conversations based on knowledge graph content corresponding to the scene information entities; receiving current problem information and identifying scene information entities of current multi-turn conversations in the current problem information; performing subgraph search operation in the knowledge graph based on a primary node, a secondary node, a tertiary node and a quaternary node in a scene information entity of current multi-turn conversation to obtain the knowledge graph corresponding to the current scene information entity, and then outputting a guidance question-answering or outputting a final answer reply based on a semantic slot rule.

In the process of implementing the embodiments of the present disclosure, it is found that at least the following problems exist in the related art:

the method does not consider the context information in multiple rounds of conversations, and in the same round of conversations, a certain relation exists among the historical questions of the user, the historical answers of the system and the current questions of the user. Because the method lacks identification and judgment of logical connection among multiple rounds of conversation information and is mainly based on the current conversation information, the generated reply information has low accuracy and even the situation of deviating from the intention of the user may occur, and finally the actual conversation experience of the user is poor.

Disclosure of Invention

The following presents a simplified summary in order to provide a basic understanding of some aspects of the disclosed embodiments. This summary is not an extensive overview nor is intended to identify key/critical elements or to delineate the scope of such embodiments but rather as a prelude to the more detailed description that is presented later.

The embodiment of the disclosure provides a method and a device for assisting a foreign language intelligent conversation, an intelligent conversation system and a storage medium, which can ensure the accuracy of generating a reply and improve the context consistency of a generated conversation so as to improve the actual conversation experience of a user.

In some embodiments, the method comprises:

acquiring user dialogue information and extracting semantic key information;

matching semantic associated information in a lower-layer common knowledge map according to the semantic key information;

acquiring semantic expansion information from an upper-layer affair knowledge graph according to the semantic key information;

and inputting the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information.

In some embodiments, the apparatus includes a processor and a memory storing program instructions, the processor being configured to, when executing the program instructions, perform the above-described method for assisting intelligent dialog in a foreign language.

In some embodiments, the intelligent dialogue system comprises the device for assisting intelligent foreign language dialogue.

In some embodiments, the storage medium stores program instructions that, when executed, perform the above-described method for assisting intelligent foreign language dialogs.

The method and the device for assisting the foreign language intelligent conversation, the intelligent conversation system and the storage medium provided by the embodiment of the disclosure can realize the following technical effects:

according to the embodiment of the disclosure, the accuracy and richness of knowledge in related fields can be ensured by importing the knowledge graph, meanwhile, the problem of data redundancy in a knowledge base is solved, and the retrieval speed of effective information is improved. By utilizing the pre-constructed lower-layer common sense knowledge map, the pre-constructed upper-layer affair knowledge map and the trained multi-turn dialogue model, the embodiment of the disclosure can realize association inquiry and effective reasoning based on user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience.

The foregoing general description and the following description are exemplary and explanatory only and are not restrictive of the application.

Drawings

One or more embodiments are illustrated by way of example in the accompanying drawings, which correspond to the accompanying drawings and not in limitation thereof, in which elements having the same reference numeral designations are shown as like elements and not in limitation thereof, and wherein:

fig. 1 is a schematic diagram of a method for assisting intelligent foreign language dialogue provided by an embodiment of the disclosure;

FIG. 2 is a schematic diagram of an event extraction model provided by an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a multi-turn dialogue model provided by an embodiment of the present disclosure;

fig. 4 is a schematic diagram of another method for assisting intelligent foreign language dialogues provided by embodiments of the present disclosure;

fig. 5 is a schematic diagram of another method for assisting intelligent foreign language dialogues provided by embodiments of the present disclosure;

fig. 6 is a schematic diagram of another method for assisting intelligent foreign language dialogues provided by embodiments of the present disclosure;

fig. 7 is a schematic diagram of an apparatus for assisting intelligent foreign language dialogues provided by the embodiments of the present disclosure.

Detailed Description

So that the manner in which the features and elements of the disclosed embodiments can be understood in detail, a more particular description of the disclosed embodiments, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. In the following description of the technology, for purposes of explanation, numerous details are set forth in order to provide a thorough understanding of the disclosed embodiments. However, one or more embodiments may be practiced without these details. In other instances, well-known structures and devices may be shown in simplified form in order to simplify the drawing.

The terms "first," "second," and the like in the description and in the claims, and the above-described drawings of embodiments of the present disclosure, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It should be understood that the data so used may be interchanged under appropriate circumstances such that embodiments of the present disclosure described herein may be made. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions.

The term "plurality" means two or more unless otherwise specified.

In the embodiment of the present disclosure, the character "/" indicates that the preceding and following objects are in an or relationship. For example, A/B represents: a or B.

The term "and/or" is an associative relationship that describes objects, meaning that three relationships may exist. For example, a and/or B, represents: a or B, or A and B.

The term "correspond" may refer to an association or binding relationship, and a corresponds to B refers to an association or binding relationship between a and B.

At present, with the improvement of education level, more and more people attach importance to foreign language teaching. To assist foreign language teaching, a number of smart question-answering systems and chatting robots have recently been introduced, which are capable of recognizing a user's question and generating a response to assist the user in learning a foreign language by matching answers to a corresponding knowledge base. However, such products have a high degree of dependence on the amount of data stored in the knowledge base, and often require a huge data set to be stored in advance to ensure the accuracy of the response, and the excessive data can seriously reduce the matching speed. Therefore, another technology is to construct a knowledge graph, wherein the knowledge graph comprises a first-level node, a second-level node, a third-level node and a fourth-level node; generating semantic slot rules for multi-turn conversations based on knowledge graph content corresponding to the scene information entities; receiving current problem information and identifying scene information entities of current multi-turn conversations in the current problem information; performing subgraph search operation in the knowledge graph based on a primary node, a secondary node, a tertiary node and a quaternary node in a scene information entity of current multi-turn conversation to obtain the knowledge graph corresponding to the current scene information entity, and then outputting a guidance question-answering or outputting a final answer reply based on a semantic slot rule. However, the method does not consider the context information in multiple rounds of conversations, and in the same round of conversations, the historical questions of the user, the historical answers of the system and the current questions of the user are in certain connection. Because the method lacks identification and judgment of logical connection among multiple rounds of conversation information and is mainly based on the current conversation information, the generated reply information has low accuracy and even the situation of deviating from the intention of the user may occur, and finally the actual conversation experience of the user is poor.

Before describing the technical solution of the present invention, the following explanation of the knowledge graph is required.

The knowledge map is a knowledge base based on binary relations, and is used for describing entities (or concepts, the concepts are abstractions of the entities, for example, the concept of 'fruit' is 'apple') and the mutual relations thereof in the real world, the basic composition unit of the knowledge map is 'entity-relation-entity' triple, and the entities are mutually connected through relations to form a network structure. Through the knowledge graph, a user can be supported to search according to subjects instead of character strings, and therefore information search on a semantic level is truly realized. The search engine based on the knowledge graph can directly feed back the structured knowledge to the user, and the user can find the knowledge which the user wants to obtain without browsing a large number of webpages.

The existing knowledge base generally takes concepts and relationships among the concepts as a core, and lacks of mining of knowledge of 'affair logic', and the affair logic (evolution rules and modes among events) is very valuable common knowledge, and the mining of the knowledge is very significant for understanding human behaviors and social development change rules. As a classic example, when a Beijing people buy a house, the house is decorated next time after buying, furniture can be bought after finishing decoration, and if a person finds a microblog on the network and says that the person buys the house, a decoration company can follow up to advertise, which is prediction. The event knowledge graph is not a knowledge base taking nouns as core nodes, but a case logic knowledge base taking events and abstract events as cores.

As shown in fig. 1, an embodiment of the present disclosure provides a method for assisting intelligent foreign language dialog, including:

s101, the processor acquires user dialogue information and extracts semantic key information.

And S102, matching semantic associated information in the lower-layer common sense knowledge graph by the processor according to the semantic key information.

S103, the processor acquires semantic expansion information from the upper-layer affair knowledge graph according to the semantic key information.

And S104, the processor inputs the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information.

By adopting the method for assisting the foreign language intelligent conversation provided by the embodiment of the disclosure, the accuracy and richness of knowledge in related fields can be ensured by introducing the knowledge map, meanwhile, the problem of data redundancy in a knowledge base is solved, and the retrieval speed of effective information is improved. By utilizing the pre-constructed lower-layer common sense knowledge map, the pre-constructed upper-layer affair knowledge map and the trained multi-turn dialogue model, the embodiment of the disclosure can realize association inquiry and effective reasoning based on the user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience.

Optionally, the processor acquires user dialogue information and extracts semantic key information, including: the processor acquires user dialogue information and analyzes semantic content from the user dialogue information; and according to the semantic content, the processor extracts semantic key information by using the trained event extraction model. Therefore, the embodiment of the disclosure can utilize the trained event extraction model to quickly identify the key information in the user dialogue information and filter out invalid information. Therefore, the relevant information can be inquired in the knowledge graph based on the intention of the user, the accuracy of reply generation is further ensured, the context consistency of the generated dialogue is improved, and the actual dialogue experience of the user is improved.

Optionally, the semantic key information includes entity information and event information. Specifically, the extracted entity information can correspond to entities and entity relationships stored in the lower common sense knowledge graph, and the extracted event information can correspond to events and event relationships stored in the upper affair knowledge graph. Therefore, according to the embodiment of the disclosure, the user can go to the lower-layer common sense knowledge graph and the upper-layer affair knowledge graph respectively to search for related information according to the entity information and the event information, so that the related inquiry and the effective reasoning can be realized based on the user dialogue information. The method and the device are beneficial to guaranteeing the accuracy of the generated response, and can improve the context consistency of the generated conversation, thereby improving the actual conversation experience of the user.

Optionally, with reference to fig. 2, an embodiment of the present disclosure provides a schematic structural diagram of an event extraction model. Specifically, the event extraction model is based on a Bi-directional Long Short-Term Memory (Bi-LSTM) layer and a Conditional Random Field (CRF) layer for architecture. The Bi-LSTM layer is formed by combining a forward LSTM layer and a backward LSTM layer, an input sequence is respectively input to the two LSTMs in a positive sequence and a negative sequence for feature extraction, and a word vector formed by splicing two output vectors (namely extracted feature vectors) is used as a subsequent feature expression of the word. Compared with a single LSTM, the Bi-LSTM can better capture bidirectional semantic dependence, so that the user intention can be grasped more accurately. The CRF layer can ensure that the final prediction result is effective by adding some constraint conditions, so that the accuracy of the extracted semantic key information is further improved. These constraints can also be learned automatically by the CRF layer when training the data.

Alternatively, the event extraction model may vector convert the input text content using a Bidirectional Encoder Representation from transforms (BERT) model from the Transformer. Among them, BERT is a Natural Language Processing (NLP) pre-training method recently proposed by google, and a general "Language understanding" model is trained on a large text corpus (e.g., wikipedia). Specifically, the input embedding for each word thereof is composed of 3 embedding, respectively, mark embedding (Token embedding), Segment embedding (Segment embedding), and Position embedding (Position embedding). Compared with the conventional Word Embedding (Word Embedding) method, the application of the BERT model not only can process the ambiguous Word problem which is difficult to solve by the Word Embedding method, but also can predict the relevance between sentences. Therefore, the user intention can be better grasped, and the semantic key information can be more accurately extracted. Furthermore, user intent often also relates to the part of speech of the word. Therefore, the event extraction model can label the part of speech of the word, and further obtain the corresponding part of speech vector. The method is favorable for further improving the accuracy of the extracted semantic key information.

Optionally, the processor matches semantic associated information in the underlying common sense knowledge-graph according to the semantic key information, including: the processor matches one or more associated entities and associated entity relations in the lower-layer common sense knowledge map according to the entity information; the processor calculates the similarity of the matched associated entities and the associated entity relation by combining historical dialogue information; and the processor determines semantic association information according to the associated entities and the similarity of the relationship of the associated entities. Thus, according to the entity information in the semantic key information, the embodiment of the disclosure can obtain the associated entity and the associated entity relationship in the lower-layer common sense knowledge graph based on the matching of the form of the triples. Optionally, the associated entity relationship may be a corresponding relationship between two entities, or may be a corresponding attribute between an entity and an attribute value. More than one associated entity and associated entity relationship is typically obtained. Specifically, in some embodiments, for an entity of a workgroup, the associated entity may be sichuan, and the corresponding associated entity relationship is a congress, that is, a triplet that constitutes "sichuan-congress-workgroup". Similarly, the associated entity may be china, and the relationship of the associated entity at this time is a city, which constitutes a triple of "china-city-capital". In addition, the associated entity may also be a numerical value, in which case the associated entity relationship is an attribute. For example, a quorum may constitute a "quorum-area-14335 square kilometers" triplet with 14335 square kilometers. And carrying out appropriate screening on the plurality of associated entities and the associated entity relationship obtained by searching. By referring to the historical dialog information of the user, the embodiment of the disclosure can better grasp the intention of the user per se according to the context information. Through the calculation of the similarity, most invalid information can be filtered out, and only the information with higher association degree is reserved as semantic association information. Therefore, the accuracy of the generated response can be ensured, and the actual conversation experience of the user can be improved.

Optionally, the processor determines semantic association information according to the similarity between the associated entities and the relationship between the associated entities, including: the processor compares the similarity of the associated entities and the relationship of the associated entities with a similarity threshold; and for the associated entities and the associated entity relations with the similarity greater than or equal to the similarity threshold, the processor determines the associated entities and the associated entity relations as semantic associated information. Therefore, by comparing the similarity with a preset threshold, the embodiment of the disclosure can filter most invalid information, and only information with higher relevance is reserved as semantic relevance information. Therefore, the accuracy of the generated response can be ensured, and the actual conversation experience of the user can be improved.

Optionally, the processor obtains semantic expansion information from the upper-layer event knowledge graph according to the semantic key information, and the semantic expansion information includes: the processor searches one or more associated events and associated event relations in the upper-layer event knowledge graph according to the event information; combining historical dialogue information, the processor calculates the possibility of the searched association event and the association event relation; and the processor determines semantic expansion information according to the associated events and the possibility of the relationship of the associated events. In this way, according to event information in the semantic key information, the embodiment of the disclosure may query the upper-layer event knowledge graph to obtain associated events and associated event relationships based on the logical relationships between the events. Optionally, the incident relationship may be one or more of a causal relationship, a conditional relationship, an inverse relationship, an order-bearing relationship, an up-down relationship, a composition relationship, and a concurrent relationship. In particular, causal relationships describe a causal consequence link in the cognitive system, where a previous event leads to a subsequent event. The condition relation describes a condition result relation in a cognitive system and is a preset and result logic. The inverse relationship often describes a mutual exclusion logic in the cognitive system, and is a true and false value logic. A time partial order relation in a cognitive system described by sequential relation is a sequential action logic. Compositional relationships, delineating the logic of whole and parts between events. The upper and lower relation describes the logic of events in the classification system. The concurrency relationship describes a symbiotic relationship of events in time, which means that one event is generated and the other event is definitely generated. Illustratively, the two events of earthquake and house collapse can form a causal relationship, and the dish buying and the cooking can form a sequential relationship. Therefore, there is usually more than one specifically obtained associated event and associated event relationship. Specifically, in some embodiments, for the event of sleeping, the associated event may be rest, and the two are in an up-down relationship. Similarly, the associated event may be eye closure, and in this case, a concurrent relationship is formed. And appropriate screening is required to be carried out on the plurality of correlation events and correlation event relations obtained by searching. By referring to the historical dialog information of the user, the embodiment of the disclosure can better grasp the intention of the user per se according to the context information. Through calculation of the possibility, most of invalid information can be filtered out, and only information with high association degree is reserved as semantic expansion information. Therefore, the context consistency of the generated conversation can be improved, and the actual conversation experience of the user can be improved.

Optionally, the processor determines semantic expansion information according to the associated event and the possibility of the relationship between the associated events, including: the processor compares the likelihood of the associated event and associated event relationship to a likelihood threshold; and determining the associated events and the associated event relations with the possibility greater than or equal to the possibility threshold value as semantic expansion information by the processor. In this way, by comparing the possibility with the preset threshold, the embodiment of the present disclosure may filter out most invalid information, and only retain information with a higher degree of association as semantic expansion information. Therefore, the context consistency of the generated conversation can be improved, and the actual conversation experience of the user can be improved.

Optionally, as shown in fig. 3, an embodiment of the present disclosure provides a schematic structural diagram of a multi-turn dialogue model. Specifically, the multi-turn dialog model includes an Embedding (Embedding) layer, an encoding (Encoder) layer, and a decoding (Decoder) layer. The embedded layer is configured to convert input semantic related information and semantic expansion information into vectors, the coding layer is configured to convert input semantic key information into vectors, and the decoding layer is configured to output a target text sequence. In this way, through the multi-turn dialog model, the embodiment of the present disclosure can ensure the accuracy of generating the response, and promote the context consistency of generating the dialog, which is beneficial to improving the actual dialog experience of the user.

Optionally, the multi-turn conversation model may also incorporate an Attention Mechanism (Attention Mechanism) layer. Thus, through weight calculation, interference of irrelevant information can be better eliminated. The accuracy of the generated response is further ensured, and the context consistency of the generated conversation is improved so as to improve the actual conversation experience of the user.

Optionally, the processor inputs the semantic key information, the semantic association information, and the semantic expansion information into the trained multi-turn dialogue model, and generates reply information corresponding to the user dialogue information, including: the processor splices the semantic associated information and the semantic expansion information to obtain semantic related information; the processor inputs the semantic related information into an embedding layer in the multi-turn dialogue model and inputs the semantic key information into a coding layer in the multi-turn dialogue model; the processor controls the multi-turn dialogue model to output reply information corresponding to the user dialogue information. In this way, with the trained multi-turn dialogue model, the embodiment of the disclosure can quickly generate the reply information corresponding to the user dialogue information, thereby realizing effective dialogue with the user. And the information input into the multi-turn dialogue model not only has semantic key information in the user dialogue information, but also comprises semantic related information after splicing processing. Because the semantic related information is derived from the lower-layer common sense knowledge graph and the upper-layer affair knowledge graph, the embodiment of the disclosure can realize related query and effective inference based on the user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience.

Optionally, the processor performs splicing processing on the semantic associated information and the semantic expansion information to obtain semantic related information, including: the processor performs entity alignment, entity disambiguation and relationship alignment on the semantic association information and the semantic expansion information to obtain semantic combination information; the processor judges whether contradiction information exists in the semantic combined information; if the semantic combination information contains contradictory information, the processor deletes the contradictory information to obtain semantic related information. Therefore, the embodiment of the disclosure can organically integrate the information inquired in the lower-layer common-sense knowledge map and the upper-layer affair knowledge map, thereby avoiding repeated input of the information and input of wrong information, and effectively improving the speed and accuracy of generating the response of the multi-turn dialogue model.

Optionally, the processor deleting contradictory information comprises: the processor distinguishes a part from the semantic relation information and a part from the semantic expansion information in the contradictory information; the processor deletes the part from the semantic expansion information in the contradictory information and reserves the part from the semantic association information in the contradictory information. In this way, when there is contradictory information in the semantic combination information, the embodiment of the present disclosure can retain common sense information with higher accuracy. Therefore, the accuracy of the semantic related information input into the multi-turn dialogue model is improved, the accuracy of the generated response is further ensured, and the improvement of the actual dialogue experience of the user is facilitated.

Optionally, after generating the reply information corresponding to the user dialog information, the method further includes: the processor controls the output module to play the reply message. Therefore, the embodiment of the disclosure can realize effective conversation with the user, and is beneficial to improving the actual conversation experience of the user.

As shown in fig. 4, another method for assisting intelligent foreign language dialog is provided in an embodiment of the present disclosure, which includes:

s401, the processor extracts knowledge data of foreign language related fields based on encyclopedia sites or vertical sites.

S402, the processor acquires entities and entity relations from the knowledge data and constructs a lower-layer common sense knowledge map.

And S403, preprocessing the knowledge data by the processor and training an event extraction model.

S404, the processor extracts the events and the event relations from the knowledge data by using the trained event extraction model, and constructs an upper-layer affair knowledge graph.

S405, the processor acquires user dialogue information and extracts semantic key information.

S406, the processor matches semantic associated information in the lower-layer common sense knowledge map according to the semantic key information.

S407, the processor acquires semantic expansion information from the upper-layer affair knowledge graph according to the semantic key information.

S408, the processor inputs the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information.

By adopting the method for assisting the foreign language intelligent conversation provided by the embodiment of the disclosure, the training of the event extraction model and the construction of the upper-layer affair knowledge graph and the lower-layer common-sense knowledge graph can be completed by extracting knowledge data on encyclopedia sites or vertical sites. The knowledge map can ensure the accuracy and richness of knowledge in related fields, simultaneously solve the problem of data redundancy in a knowledge base and improve the retrieval speed of effective information. By utilizing the pre-constructed lower-layer common sense knowledge map, the pre-constructed upper-layer affair knowledge map and the trained multi-turn dialogue model, the embodiment of the disclosure can realize association inquiry and effective reasoning based on user dialogue information. Thereby ensuring the accuracy of the generated response and promoting the contextual continuity of the generated dialog to improve the user's actual dialog experience.

Optionally, the lower common sense knowledge map and the upper affairs knowledge map together form a double-layer knowledge map. Specifically, the upper-layer event knowledge graph is stored by taking events and event relationships as units, and the entity is an important component in the events. The entities in the event can be linked to corresponding portions of the underlying common sense knowledge-graph by entity identification and entity linking techniques. The double-layer knowledge graph can complete the linkage of the entity and the entity relationship, the event and the event relationship and the entity and the event, thereby further expanding the knowledge base and improving the overall relation between information.

Optionally, the processor constructing the underlying common sense knowledge graph comprises: the processor carries out entity alignment, entity disambiguation and relationship alignment processing on the obtained entity and entity relationship; and integrating the processed entities and entity relations into the lower-layer common sense knowledge map by the processor through entity linkage. The entity alignment aims to judge whether entities of a plurality of different information sources point to the same object in the real world or not. If a plurality of entities represent the same object, an alignment relation is constructed among the entities, and meanwhile information contained in the entities is fused and aggregated. The entity disambiguation is to eliminate the ambiguity phenomenon of word ambiguity according to the context information. The relationship alignment is to determine whether multiple relationships can represent the same relationship, and perform information fusion on relationships with different sources or names but the same representation, thereby obtaining richer and more accurate information. Therefore, the embodiment of the disclosure can quickly establish the lower-layer common sense knowledge map, and simultaneously improves the richness and accuracy of stored knowledge, thereby ensuring the quality of the lower-layer common sense knowledge map.

Optionally, the processor constructing the upper-layer affairs knowledge graph comprises: the processor carries out event alignment, event disambiguation and relationship alignment processing on the acquired events and event relationships; and through event linkage, the processor integrates the processed events and event relations into an upper-layer event knowledge graph. The event alignment aims to judge whether events of a plurality of different information sources point to the same thing in the real world or not. If a plurality of events represent the same thing, an alignment relation is constructed among the events, and information contained in the events is fused and aggregated. Event disambiguation is the phenomenon of disambiguation of one sentence based on context information. The relationship alignment is to determine whether multiple relationships can represent the same relationship, and perform information fusion on relationships with different sources or names but the same representation, thereby obtaining richer and more accurate information. Therefore, the embodiment of the disclosure can quickly establish the upper-layer affair knowledge graph, and meanwhile, the richness and the accuracy of the stored knowledge are improved, so that the quality of the upper-layer affair knowledge graph is ensured.

Optionally, the processor preprocesses the knowledge data and trains the event extraction model, including: the processor calibrates part of data in the knowledge data to obtain calibrated knowledge data; the processor inputs the calibrated knowledge data and the uncalibrated knowledge data into the event extraction model, and trains the event extraction model through a semi-supervised learning method. Therefore, the event extraction model can be trained by using a small amount of labeled samples and a large amount of unlabeled samples, so that the dependence of the event extraction model on labeled data is greatly reduced. Meanwhile, the distribution of the unlabeled data can also provide a lot of valuable information, so that the method has guiding significance on model iteration and is beneficial to improving the performance of the event extraction model.

Optionally, after the processor extracts knowledge data of the foreign language related domain based on the encyclopedia site or the vertical site, the method further includes: the processor performs duplicate removal processing on the knowledge data; and the processor matches the knowledge data in the sensitive information database and removes the knowledge data containing the sensitive information. Therefore, before the knowledge graph is constructed and the event extraction model is trained, the embodiment of the disclosure can carry out the preprocessing of removing the duplication and the sensitive words on the knowledge data, thereby ensuring the legality of the extracted information. Thereby being beneficial to ensuring the accuracy of the generated reply.

As shown in fig. 5, another method for assisting intelligent foreign language dialog is provided in an embodiment of the present disclosure, including:

s501, the processor extracts knowledge data of foreign language related fields based on encyclopedia sites or vertical sites.

And S502, the processor acquires entities and entity relations from the knowledge data and constructs a lower-layer common sense knowledge map.

S503, the processor preprocesses the knowledge data and trains the event extraction model.

S504, the processor extracts the events and the event relations from the knowledge data by using the trained event extraction model, and constructs an upper-layer affair knowledge graph.

S505, the processor judges whether the extracted event and the event relation contain common sense information.

S506, if the extracted events and event relations comprise common sense information, the processor extracts entities and entity relations from the common sense information and collects the entities and entity relations into a lower common sense knowledge map.

And S507, the processor acquires the user dialogue information and extracts semantic key information.

And S508, matching the semantic associated information in the lower-layer common sense knowledge graph by the processor according to the semantic key information.

S509, the processor acquires semantic expansion information from the upper-layer matter knowledge graph according to the semantic key information.

S510, the processor inputs the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information.

By adopting the method for assisting the foreign language intelligent conversation provided by the embodiment of the disclosure, the training of the event extraction model and the construction of the upper-layer affair knowledge graph and the lower-layer common-sense knowledge graph can be completed by extracting knowledge data on encyclopedia sites or vertical sites. In addition, the embodiment of the disclosure can further extract the entity and the entity relationship by utilizing the event and the event relationship extracted by the event extraction model, thereby enriching and perfecting the information of the lower-layer common sense knowledge map. The knowledge map can ensure the accuracy and richness of knowledge in related fields, simultaneously solve the problem of data redundancy in a knowledge base and improve the retrieval speed of effective information. By utilizing the pre-constructed lower-layer common sense knowledge map, the pre-constructed upper-layer affair knowledge map and the trained multi-turn dialogue model, the embodiment of the disclosure can realize association inquiry and effective reasoning based on user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience.

As shown in fig. 6, another method for assisting intelligent foreign language dialog is provided in an embodiment of the present disclosure, including:

s601, the processor extracts corpus data related to foreign language conversation based on the teaching material question bank or the social network site.

S602, the processor preprocesses the speech data and trains a multi-round dialogue model.

S603, the processor acquires the user dialogue information and extracts the semantic key information.

S604, the processor matches semantic associated information in the lower-layer common sense knowledge map according to the semantic key information.

And S605, the processor acquires semantic expansion information from the upper-layer affair knowledge graph according to the semantic key information.

And S606, the processor inputs the semantic key information, the semantic association information and the semantic expansion information into the trained multi-turn dialogue model to generate reply information corresponding to the user dialogue information.

By adopting the method for assisting the foreign language intelligent conversation provided by the embodiment of the disclosure, the training of a multi-round conversation model can be completed by extracting the corpus data on the teaching material library or the social network site. The accuracy and richness of knowledge in related fields can be ensured by introducing the knowledge graph, the problem of data redundancy in a knowledge base is solved, and the retrieval speed of effective information is improved. By utilizing the pre-constructed lower-layer common sense knowledge map, the pre-constructed upper-layer affair knowledge map and the trained multi-turn dialogue model, the embodiment of the disclosure can realize association inquiry and effective reasoning based on user dialogue information. Thereby ensuring the accuracy of generating the response and improving the contextual consistency of the generated dialog to improve the user's actual dialog experience.

Optionally, the processor preprocesses the speech data, trains a multi-turn dialogue model, and includes: the processor calibrates part of data in the corpus data to obtain calibrated corpus data; the processor inputs the calibrated corpus data and the uncalibrated corpus data into the multi-turn dialogue model, and trains the multi-turn dialogue model through a semi-supervised learning method. In this way, the embodiment of the present disclosure can train a multi-round dialogue model using a small number of labeled samples and a large number of unlabeled samples, thereby greatly reducing the dependence of the multi-round dialogue model on labeled data. Meanwhile, the distribution of the unlabeled data can also provide a lot of valuable information, so that the method has guiding significance on model iteration and is beneficial to improving the performance of a multi-turn dialogue model.

Optionally, after the processor extracts corpus data related to the foreign language conversation based on the teaching material question bank or the social network site, the method further comprises: the processor performs duplicate removal processing on the material data; and the processor matches the corpus data in the sensitive information database, and removes the corpus data containing the sensitive information. Therefore, before training a multi-turn dialogue model, the embodiment of the disclosure can perform preprocessing of removing the duplication and the sensitive words on the corpus data, thereby ensuring the legality of extracting information. Thereby being beneficial to ensuring the accuracy of the generated reply.

As shown in fig. 7, an apparatus for assisting intelligent foreign language dialogue according to an embodiment of the present disclosure includes a processor 701 and a memory 702. Optionally, the apparatus may also include a Communication Interface 703 and a bus 704. The processor 701, the communication interface 703 and the memory 702 may communicate with each other through a bus 704. Communication interface 703 may be used for the transfer of information. The processor 701 may invoke logic instructions in the memory 702 to perform the method for assisting a foreign language intelligent dialog of the above-described embodiments.

Furthermore, the logic instructions in the memory 702 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product.

The memory 702 is a computer-readable storage medium, and can be used for storing software programs, computer-executable programs, such as program instructions/modules corresponding to the methods in the embodiments of the present disclosure. The processor 701 executes functional applications and data processing by executing program instructions/modules stored in the memory 702, that is, implements the method for assisting the foreign language intelligent dialogue in the above-described embodiment.

The memory 702 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal device, and the like. In addition, the memory 702 may include high speed random access memory, and may also include non-volatile memory.

The embodiment of the disclosure provides an intelligent dialogue system, which comprises the device for assisting the intelligent foreign language dialogue.

Optionally, in some embodiments, the intelligent dialog system further comprises an input module and an output module. Wherein the input module is configured to obtain user dialog information. The output module is configured to play the reply message. Therefore, the intelligent dialogue system can establish effective dialogue with the user, and actual dialogue experience of the user is guaranteed.

Alternatively, the input module may be a microphone so that the user dialog information can be continuously acquired.

Alternatively, the output module may be a speaker, so that the reply message can be played in time.

The disclosed embodiments provide a storage medium storing computer-executable instructions that, when executed, perform the above-described method for assisting intelligent foreign language dialogues.

The disclosed embodiments provide a computer program product comprising a computer program stored on a storage medium, the computer program comprising program instructions which, when executed by a computer, cause the computer to perform the above-described method for assisting intelligent dialogues of a foreign language.

The storage medium described above may be a transitory computer-readable storage medium or a non-transitory computer-readable storage medium.

The technical solution of the embodiments of the present disclosure may be embodied in the form of a software product, where the computer software product is stored in a storage medium and includes one or more instructions to enable a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method of the embodiments of the present disclosure. And the aforementioned storage medium may be a non-transitory storage medium comprising: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes, and may also be a transient storage medium.

The above description and drawings sufficiently illustrate embodiments of the disclosure to enable those skilled in the art to practice them. Other embodiments may incorporate structural, logical, electrical, process, and other changes. The examples merely typify possible variations. Individual components and functions are optional unless explicitly required, and the sequence of operations may vary. Portions and features of some embodiments may be included in or substituted for those of others. Furthermore, the words used in the specification are words of description only and are not intended to limit the claims. As used in the description of the embodiments and the claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Similarly, the term "and/or" as used in this application is meant to encompass any and all possible combinations of one or more of the associated listed. Furthermore, the terms "comprises" and/or "comprising," when used in this application, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. Without further limitation, an element defined by the phrase "comprising an …" does not exclude the presence of other like elements in a process, method or apparatus that comprises the element. In this document, each embodiment may be described with emphasis on differences from other embodiments, and the same and similar parts between the respective embodiments may be referred to each other. For methods, products, etc. of the embodiment disclosures, reference may be made to the description of the method section for relevance if it corresponds to the method section of the embodiment disclosure.

Those of skill in the art would appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software may depend upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the disclosed embodiments. It can be clearly understood by the skilled person that, for convenience and brevity of description, the specific working processes of the system, the apparatus and the unit described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, apparatuses, etc.) may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units may be merely a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form. The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to implement the present embodiment. In addition, functional units in the embodiments of the present disclosure may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. In the description corresponding to the flowcharts and block diagrams in the figures, operations or steps corresponding to different blocks may also occur in different orders than disclosed in the description, and sometimes there is no specific order between the different operations or steps. For example, two sequential operations or steps may in fact be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. Each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Claims

1. A method for facilitating intelligent foreign language dialogues, comprising:

acquiring user dialogue information and extracting semantic key information;

2. The method according to claim 1, wherein the lower-level common sense knowledge-graph and the upper-level affairs knowledge-graph together form a double-level knowledge-graph, and obtaining the double-level knowledge-graph comprises:

extracting knowledge data of foreign language related fields based on encyclopedia sites or vertical sites;

acquiring entities and entity relations from knowledge data, and constructing a lower-layer common sense knowledge map;

preprocessing knowledge data and training an event extraction model;

and extracting a model from the knowledge data by using the trained events to extract events and event relations, and constructing an upper-layer affair knowledge graph.

3. The method of claim 2, wherein preprocessing the knowledge data and training the event extraction model comprises:

calibrating partial data in the knowledge data to obtain calibrated knowledge data;

inputting the calibrated knowledge data and the uncalibrated knowledge data into an event extraction model, and training the event extraction model by a semi-supervised learning method.

4. The method of claim 2, wherein after the building of the upper-level situational knowledge graph, the method further comprises:

judging whether the extracted events and event relations contain common sense information or not;

if the extracted events and event relations comprise common sense information, entities and entity relations are extracted from the common sense information and are gathered to a lower-layer common sense knowledge map.

5. The method of claim 1, wherein obtaining the trained multi-turn conversation model comprises:

extracting corpus data related to foreign language conversation based on a teaching material question bank or a social network site;

and preprocessing the corpus data and training a multi-round dialogue model.

6. The method of claim 5, wherein preprocessing the corpus data and training a multi-turn dialogue model comprises:

calibrating part of data in the corpus data to obtain calibrated corpus data;

and inputting the calibrated corpus data and the uncalibrated corpus data into a multi-turn dialogue model, and training the multi-turn dialogue model by a semi-supervised learning method.

7. The method according to any one of claims 1 to 6, wherein the obtaining of the user dialogue information and the extracting of the semantic key information comprise:

acquiring user dialogue information, and analyzing semantic content from the user dialogue information;

extracting semantic key information by using a trained event extraction model according to semantic content;

the semantic key information comprises entity information and event information.

8. The method according to claim 7, wherein the matching semantic associated information in the underlying common sense knowledge-graph according to the semantic key information comprises:

matching one or more associated entities and associated entity relations in a lower-layer common sense knowledge graph according to entity information;

calculating the similarity of the matched associated entities and the associated entity relationship by combining historical dialogue information;

and determining semantic associated information according to the associated entities and the similarity of the relationship of the associated entities.

9. The method according to claim 7, wherein the obtaining semantic expansion information from the upper-layer event knowledge graph according to the semantic key information comprises:

according to the event information, searching one or more associated events and associated event relations in an upper-layer event knowledge graph;

carrying out probability calculation on the searched associated events and the relationship of the associated events by combining historical conversation information;

and determining semantic expansion information according to the associated events and the possibility of the relationship of the associated events.

10. An apparatus for assisting intelligent foreign language dialogues, comprising a processor and a memory having stored thereon program instructions, wherein the processor is configured to execute the method for assisting intelligent foreign language dialogues as claimed in any one of claims 1 to 9 when executing the program instructions.

11. An intelligent dialog system, characterized in that it comprises means for assisting intelligent dialogs of foreign languages as claimed in claim 10.

12. A storage medium storing program instructions, characterized in that said program instructions, when executed, perform a method for assisting intelligent foreign language dialogues according to any of claims 1 to 9.