CN114201595A

CN114201595A - Sentence recommendation method and device in conversation, storage medium and electronic equipment

Info

Publication number: CN114201595A
Application number: CN202111518006.4A
Authority: CN
Inventors: 黄钰瑶; 冯伟; 武晓飞
Original assignee: Shell Housing Network Beijing Information Technology Co Ltd
Current assignee: Shell Housing Network Beijing Information Technology Co Ltd
Priority date: 2021-12-13
Filing date: 2021-12-13
Publication date: 2022-03-18

Abstract

The embodiment of the disclosure discloses a sentence recommendation method and device in a conversation, a storage medium and an electronic device, wherein the method comprises the following steps: determining current intention information corresponding to a first sentence sent by a first dialog body in a current dialog, and determining current action information corresponding to a second dialog body in the current dialog based on the current intention information; determining a second set of slots based on at least one historical session between a first session subject and a second session subject prior to the current session; obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; determining n target second sentences from the set of predicted second sentences as conversational recommendations for the second conversational partner. The target second sentence obtained by the embodiment is more suitable for the current scene, and the personalized sentence recommendation is realized by combining the dialogue style of the second main body.

Description

Sentence recommendation method and device in conversation, storage medium and electronic equipment

Technical Field

The present disclosure relates to the field of information interaction technologies, and in particular, to a method and an apparatus for recommending sentences in a conversation, a storage medium, and an electronic device.

Background

With the development of science and technology, more and more intelligent auxiliary devices appear in life and work of people, wherein a conversation robot is a common robot. The dialogue robot can be used in customer service, after the inquiry intention of the user is identified, a corresponding answer template is called through the intention, and the answer is generated by combining the data of the inquiry and answer target; however, the answer template in the prior art is too single, does not consider a large amount of information contained in the chat text, and cannot generate different answer styles.

Disclosure of Invention

The present disclosure is proposed to solve the above technical problems. The embodiment of the disclosure provides a sentence recommendation method and device in a conversation, a storage medium and an electronic device.

According to an aspect of an embodiment of the present disclosure, there is provided a sentence recommendation method in a dialog, including:

determining current intention information corresponding to a first sentence sent by a first dialog body in a current dialog, and determining current action information corresponding to a second dialog body in the current dialog based on the current intention information;

determining a second set of slots based on at least one historical session between the first session subject and the second session subject prior to the current session; wherein the second slot position set comprises at least one slot position information;

obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; wherein the set of predicted second sentences includes at least one predicted second sentence;

determining n target second sentences from the set of predicted second sentences as conversation recommendations for the second conversation subject; wherein n is an integer of 1 or more.

Optionally, the determining current intention information corresponding to a first sentence, which is uttered by a first dialog body in a current dialog, and determining current action information corresponding to a second dialog body in the current dialog based on the current intention information includes:

receiving the first sentence sent by the first conversation main body in the current conversation;

performing intention recognition on the first sentence by using a first network model to obtain current intention information corresponding to the first sentence;

and determining the predicted action information corresponding to the current intention information by utilizing a second network model, and taking the predicted action information as the current action information corresponding to the second dialogue main body.

Optionally, the determining a second set of slots based on at least one historical session between the first session master and the second session master prior to the current session comprises:

acquiring at least one historical first sentence and at least one historical second sentence which are included in at least one historical conversation; wherein each history session comprises a session body transfer, and the session body comprises the first session body and the second session body;

determining a first slot position set corresponding to the at least one historical first sentence and the at least one historical second sentence by utilizing a third network model; wherein the first slot set comprises at least one slot information;

determining at least one slot position information from the first slot position set to form a second slot position set based on the current intention information and the current action information.

Optionally, the determining at least one slot information from the first slot set to constitute a second slot set based on the current intent information and the current action information comprises:

determining a prediction slot position set corresponding to the current conversation based on the current intention information and the current action information by utilizing a fourth network model; the prediction slot position set comprises at least one slot position information;

determining the second set of slots based on an intersection of the first set of slots and the predicted set of slots.

Optionally, the obtaining a predicted second sentence set based on the current motion information, the second slot set, and the portrait information corresponding to the second main body includes:

mapping the current motion information, the second slot position set and the portrait information into vector expressions respectively to obtain a motion vector, a slot position vector and a portrait vector;

processing the motion vector, the slot position vector and the portrait vector to obtain a combined vector;

processing the combined vector based on the trained generative model to obtain at least one predicted second sentence in the set of predicted second sentences.

Optionally, the obtaining at least one predicted second sentence based on the current motion information, the second slot set, and the portrait information corresponding to the second main body includes:

processing the combined vector based on the trained classification model, and determining at least one template sentence from a preset template set as at least one predicted second sentence in the predicted second sentence set; and the preset template set comprises a plurality of template sentences.

Optionally, the determining n target second sentences from the predicted second sentence set as the conversation recommendation of the second conversation subject includes:

forming at least one predicted conversation based on the first sentence and at least one of the predicted second sentences included in the set of predicted second sentences;

scoring each of the at least one prediction session by using a semantic matching model to obtain at least one evaluation score; wherein each said rating score corresponds to a prediction session;

determining the n target second sentences as the conversation recommendations of the second conversation subject based on each of the at least one rating scores.

Optionally, the determining the n target second sentences as the conversation recommendation of the second conversation subject based on each of the at least one rating score includes:

sequencing the at least one prediction session according to the magnitude of the at least one evaluation score to obtain a session sequence;

taking the predicted second sentences in the first n predicted conversations from the conversation sequence in sequence as target second sentences to obtain n target second sentences;

and taking the n target second sentences as the conversation recommendation of the second conversation body.

According to another aspect of the embodiments of the present disclosure, there is provided a sentence recommendation apparatus in a dialog, including:

the information determining module is used for determining current intention information corresponding to a first sentence sent by a first dialog main body in a current dialog, and determining current action information corresponding to a second dialog main body in the current dialog based on the current intention information;

a slot determination module to determine a second set of slots based on at least one historical session between the first session master and the second session master prior to the current session; wherein the second slot position set comprises at least one slot position information;

the initial prediction module is used for obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; wherein the set of predicted second sentences includes at least one predicted second sentence;

a conversation recommendation module, configured to determine n target second sentences from the predicted second sentence set as conversation recommendations of the second conversation subject; wherein n is an integer of 1 or more.

Optionally, the information determining module is specifically configured to receive the first sentence, which is sent by the first dialog body in the current dialog; performing intention recognition on the first sentence by using a first network model to obtain current intention information corresponding to the first sentence; and determining the predicted action information corresponding to the current intention information by utilizing a second network model, and taking the predicted action information as the current action information corresponding to the second dialogue main body.

Optionally, the slot determining module includes:

the history conversation unit is used for acquiring at least one history first sentence and at least one history second sentence which are included in at least one history conversation; wherein each history session comprises a session body transfer, and the session body comprises the first session body and the second session body;

a slot extracting unit, configured to determine, by using a third network model, a first slot set corresponding to the at least one historical first sentence and corresponding to the at least one historical second sentence; wherein the first slot set comprises at least one slot information;

a slot position set unit, configured to determine at least one slot position information from the first slot position set to constitute a second slot position set based on the current intention information and the current action information.

Optionally, the slot aggregation unit is specifically configured to determine, by using a fourth network model, a predicted slot aggregation corresponding to the current session based on the current intention information and the current action information; the prediction slot position set comprises at least one slot position information; determining the second set of slots based on an intersection of the first set of slots and the predicted set of slots.

Optionally, the initial prediction module is specifically configured to map the current motion information, the second slot position set, and the portrait information into vector expressions respectively, so as to obtain a motion vector, a slot position vector, and a portrait vector; processing the motion vector, the slot position vector and the portrait vector to obtain a combined vector; processing the combined vector based on the trained generative model to obtain at least one predicted second sentence in the set of predicted second sentences.

Optionally, the initial prediction module is specifically configured to map the current motion information, the second slot position set, and the portrait information into vector expressions respectively, so as to obtain a motion vector, a slot position vector, and a portrait vector; processing the motion vector, the slot position vector and the portrait vector to obtain a combined vector; processing the combined vector based on the trained classification model, and determining at least one template sentence from a preset template set as at least one predicted second sentence in the predicted second sentence set; and the preset template set comprises a plurality of template sentences.

Optionally, the session recommendation module includes:

a predicted conversation unit configured to constitute at least one predicted conversation based on the first sentence and at least one of the predicted second sentences included in the set of predicted second sentences;

the scoring unit is used for scoring each prediction session in the at least one prediction session by utilizing a semantic matching model to obtain at least one evaluation score; wherein each said rating score corresponds to a prediction session;

a target determination unit, configured to determine, based on each of the at least one rating score, the n target second sentences as a conversation recommendation of the second conversation subject.

Optionally, the target determining unit is specifically configured to sort the at least one prediction session according to the magnitude of the at least one evaluation score, so as to obtain a session sequence; taking the predicted second sentences in the first n predicted conversations from the conversation sequence in sequence as target second sentences to obtain n target second sentences; and taking the n target second sentences as the conversation recommendation of the second conversation body.

According to still another aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium storing a computer program for executing the sentence recommendation method in a dialog according to any of the embodiments.

According to still another aspect of the embodiments of the present disclosure, there is provided an electronic apparatus including:

a processor;

a memory for storing the processor-executable instructions;

the processor is configured to read the executable instruction from the memory and execute the instruction to implement the sentence recommendation method in dialog according to any of the embodiments.

Based on the sentence recommendation method and device in the dialog, the storage medium and the electronic device provided by the above embodiment of the present disclosure, current intention information corresponding to a first sentence sent by a first dialog subject in a current dialog is determined, and current action information corresponding to a second dialog subject in the current dialog is determined based on the current intention information; determining a second set of slots based on at least one historical session between the first session subject and the second session subject prior to the current session; obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; determining n target second sentences from the set of predicted second sentences as conversation recommendations for the second conversation subject; wherein n is an integer of 1 or more. In the embodiment, the second slot position set is determined by combining the historical conversation, so that the information in the conversation context is combined into the current conversation, the obtained target second sentence is more consistent with the current scene, the dialogue style of the second main body is combined by combining the portrait information of the second main body, personalized sentence recommendation is realized, the probability of selecting the recommended target second sentence is improved, and the conversation efficiency is further improved.

The technical solution of the present disclosure is further described in detail by the accompanying drawings and examples.

Drawings

The above and other objects, features and advantages of the present disclosure will become more apparent by describing in more detail embodiments of the present disclosure with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the principles of the disclosure and not to limit the disclosure. In the drawings, like reference numbers generally represent like parts or steps.

Fig. 1 is a flowchart illustrating a sentence recommendation method in a dialog according to an exemplary embodiment of the present disclosure.

FIG. 2 is a schematic flow chart of step 102 in the embodiment shown in FIG. 1 of the present disclosure.

Fig. 3 is a schematic flow chart of step 104 in the embodiment shown in fig. 1 of the present disclosure.

Fig. 4 is a schematic flow chart of step 106 in the embodiment shown in fig. 1 of the present disclosure.

Fig. 5 is another schematic flow chart of step 106 in the embodiment shown in fig. 1 of the present disclosure.

Fig. 6 is a schematic flow chart of step 108 in the embodiment shown in fig. 1 of the present disclosure.

Fig. 7 is a schematic structural diagram of a sentence recommendation device in a dialog according to an exemplary embodiment of the present disclosure.

Fig. 8 is a block diagram of an electronic device provided in an exemplary embodiment of the present disclosure.

Detailed Description

Hereinafter, example embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. It is to be understood that the described embodiments are merely a subset of the embodiments of the present disclosure and not all embodiments of the present disclosure, with the understanding that the present disclosure is not limited to the example embodiments described herein.

It should be noted that: the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

It will be understood by those of skill in the art that the terms "first," "second," and the like in the embodiments of the present disclosure are used merely to distinguish one element from another, and are not intended to imply any particular technical meaning, nor is the necessary logical order between them.

It is also understood that in embodiments of the present disclosure, "a plurality" may refer to two or more and "at least one" may refer to one, two or more.

It is also to be understood that any reference to any component, data, or structure in the embodiments of the disclosure, may be generally understood as one or more, unless explicitly defined otherwise or stated otherwise.

In addition, the term "and/or" in the present disclosure is only one kind of association relationship describing an associated object, and means that three kinds of relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in the present disclosure generally indicates that the former and latter associated objects are in an "or" relationship. The data referred to in this disclosure may include unstructured data, such as text, images, video, etc., as well as structured data.

It should also be understood that the description of the various embodiments of the present disclosure emphasizes the differences between the various embodiments, and the same or similar parts may be referred to each other, so that the descriptions thereof are omitted for brevity.

Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

The disclosed embodiments may be applied to electronic devices such as terminal devices, computer systems, servers, etc., which are operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well known terminal devices, computing systems, environments, and/or configurations that may be suitable for use with electronic devices, such as terminal devices, computer systems, servers, and the like, include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, microprocessor-based systems, set top boxes, programmable consumer electronics, network pcs, minicomputer systems, mainframe computer systems, distributed cloud computing environments that include any of the above systems, and the like.

Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The computer system/server may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.

Exemplary method

Fig. 1 is a flowchart illustrating a sentence recommendation method in a dialog according to an exemplary embodiment of the present disclosure. The embodiment can be applied to an electronic device, as shown in fig. 1, and includes the following steps:

step 102, determining current intention information corresponding to a first sentence sent by a first dialog body in a current dialog, and determining current action information corresponding to a second dialog body in the current dialog based on the current intention information.

Alternatively, the present embodiment may be applied to various different scenarios, for example, a real estate domain, etc., and the first dialog principal may be a user, and the second dialog principal may be a broker; the current intention information represents the intention of the first dialog body embodied in the first sentence, for example, in the house property field, the intention information may include, but is not limited to: price, area, floor, etc.; and the current action information represents a feedback action to the first sentence, such as an answer, a question, a negation, an acknowledgement, and the like.

Step 104, determining a second slot set based on at least one historical session between the first session subject and the second session subject prior to the current session.

The historical conversation is a conversation between a first conversation main body and a second conversation main body before the current conversation; the second slot set includes at least one slot information.

In this embodiment, a conversation only includes a main body transfer once, for example, a first conversation main body sends out a first sentence composed of at least one sentence of text, and a second conversation main body sends out a second sentence composed of at least one text, so as to realize a main body transfer once, when the first conversation main body sends out a sentence again, a second main body transfer occurs, and at this time, a next conversation is entered; the slot position information in this embodiment indicates specific numerical information, result information, and the like corresponding to the intention; for example, the intention is a price, and a specific price value is slot position information; whether the intention is to be street or not, whether the intention is slot position information or not and the like; the slot information corresponds to information of a target object exchanged in the session, for example, the target object is a house source, and the slot information may be extracted from information data of the house source.

And 106, obtaining a predicted second sentence set based on the current action information, the second slot position set and the image information corresponding to the second main body.

Wherein the set of predicted second sentences includes at least one predicted second sentence.

In this embodiment, the portrait information of the second subject is a conversation style, a conversation feature, and the like of the second subject determined based on the historical conversation data of the second subject, for example, the conversation style is short or detailed; the embodiment can generate the predicted second sentence more conforming to the conversation style of the second main body by combining the portrait information, thereby improving the probability of selecting the sentence.

And step 108, determining n target second sentences from the predicted second sentence set as conversation recommendations of a second conversation subject.

Wherein n is an integer of 1 or more.

In this embodiment, it is predicted that the number of second sentences is large, but the second sentences cannot be recommended for the second subject, that is, the recommendation efficiency is reduced, the subjective feeling of the second subject is influenced, and one of the second sentences needs to be selected from a plurality of recommendations to influence the conversation efficiency; in this embodiment, only n target second sentences are recommended for the second main body, so as to implement efficient and fast conversation recommendation, where a specific value of n may be set according to an actual scene, for example, n is 3.

In the sentence recommendation method in a dialog provided by the above embodiment of the present disclosure, current intention information corresponding to a first sentence sent by a first dialog body in a current dialog is determined, and current action information corresponding to a second dialog body in the current dialog is determined based on the current intention information; determining a second set of slots based on at least one historical session between the first session subject and the second session subject prior to the current session; obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; determining n target second sentences from the set of predicted second sentences as conversation recommendations for the second conversation subject; wherein n is an integer of 1 or more. In the embodiment, the second slot position set is determined by combining the historical conversation, so that the information in the conversation context is combined into the current conversation, the obtained target second sentence is more consistent with the current scene, the dialogue style of the second main body is combined by combining the portrait information of the second main body, personalized sentence recommendation is realized, the probability of selecting the recommended target second sentence is improved, and the conversation efficiency is further improved.

As shown in fig. 2, based on the embodiment shown in fig. 1, step 102 may include the following steps:

step 1021, receiving a first sentence sent by the first dialog body in the current dialog.

Step 1022, performing intent recognition on the first sentence by using the first network model, so as to obtain current intent information corresponding to the first sentence.

In this embodiment, the first network model may be a neural network model (e.g., a classification model, etc.) with an arbitrary structure, and the first network model is trained through a large number of sentences with known intention information to realize intention recognition on the sentences input into the first network model; in the training process, sentences are used as input samples, and known intention information is used as supervision information.

And step 1023, determining the predicted action information corresponding to the current intention information by using the second network model, and taking the predicted action information as the current action information corresponding to the second dialogue main body.

Alternatively, the second network model may be an arbitrary structural neural network model (e.g., a classification model, etc.), which is trained by a large amount of session information of known intention information and action information (the action information may be determined by labeling, for example, whether it is an answer action, in the housing estate domain, whether a broker answers a question of a user, the answer is labeled 1, otherwise, it is labeled 0, etc.), so as to perform action information prediction on the intention information input to the second network model; in the training process, intention information is used as an input sample, and known action information is used as supervision information.

In the embodiment, the action that the second main body may feed back the current intention information is predicted through the recognized current intention information, and corresponding slot position information conforming to the question and answer logic is acquired by combining the predicted current action information, so that sentence recommendation conforming to the logic is realized.

As shown in fig. 3, based on the embodiment shown in fig. 1, step 104 may include the following steps:

step 1041, obtaining at least one historical first sentence and at least one historical second sentence included in at least one historical conversation.

Wherein, each historical conversation comprises a conversation body transfer, and the conversation body comprises a first conversation body and a second conversation body.

Step 1042, determining a first slot set corresponding to the at least one historical first sentence and the at least one historical second sentence using the third network model.

Wherein the first slot set includes at least one slot information.

Optionally, the third network model may be a neural network model of any structure, and the third network model is trained by a large number of sentences of which slot position information is known, so as to extract slot position information in the sentences; in the training process, sentences are used as input samples, and known slot position information is used as supervision information.

Step 1043, determining at least one slot position information from the first slot position set to form a second slot position set based on the current intention information and the current action information.

In this embodiment, existing slot information that may appear in a sentence answered by the second dialog body in the current conversation is determined from the first slot set corresponding to the historical conversation by combining the current intention information and the current action information, and after the second slot set is determined, the corresponding slot information may be directly obtained from the first slot set, without calling from attribute data corresponding to the target object of the conversation, so that the efficiency of determining the second slot set is improved.

Optionally, in this embodiment, the step 1043 may include:

and determining a prediction slot position set corresponding to the current conversation by utilizing a fourth network model based on the current intention information and the current action information.

And the prediction slot position set comprises at least one slot position information.

Optionally, the fourth network model may be a neural network of any structure, and a joint distribution among [ actions, intentions, slots ] is learned by using known historical intention information, historical action information, and corresponding historical slot information, so that when an action and an intention are fixed, a predicted slot set of slot information that may appear in a sentence may be output based on the fourth network model.

A second set of slots is determined based on an intersection of the first set of slots and the set of predicted slots.

In this embodiment, the second slot position set is determined based on the intersection of the predicted slot position set predicted by the fourth network model and the first slot position set, that is, it is ensured that the slot position information in the second slot position set is the slot position information that may appear in the answer of the current session, and it is ensured that the slot position information in the second slot position set is in the first slot position set, so that the process of reacquiring the slot position information is reduced, the determination efficiency of the second slot position set is improved, and further the efficiency of sentence recommendation is improved.

As shown in fig. 4, based on the embodiment shown in fig. 1, step 106 may include the following steps:

step 1061, mapping the current motion information, the second slot position set, and the portrait information into vector expressions, respectively, to obtain a motion vector, a slot position vector, and a portrait vector.

Optionally, the current motion information, the second slot position set, and the portrait information may be encoded in any existing encoding manner as an expression in a vector form, for example, the motion information may be encoded as a vector of a total length of motion types, where each element in the vector corresponds to one motion type, and when the number of motion types is 4, the current motion information may be encoded as 0010, where 1 is the corresponding motion type; the embodiment does not limit the specific coding mode, and only needs to distinguish different actions, slot positions and images through vectors; in addition, in this embodiment, the identification number (used to distinguish different second session bodies) corresponding to the second session body may be encoded into a vector expression, so as to obtain an identification number vector.

Step 1062, process the motion vector, slot position vector and image vector to obtain a combined vector.

Optionally, the manner of obtaining the combined vector may be to perform vector splicing processing and/or vector addition processing on the motion vector, the slot vector, and the image vector, where the splicing processing may be dimension splicing, for example, the motion vector, the slot vector, and the image vector are 10-dimensional vectors respectively, and the combined vector obtained after splicing is a 30-dimensional vector; of course, when an identification number vector is included, the identification number vector is also spliced at the time of the splicing process.

Step 1063, processing the combined vector based on the trained generative model to obtain at least one predicted second sentence in the predicted second sentence set.

In this embodiment, the generative model may be a generative model of any structure (e.g., GPT-2 model, etc.), the generative model is trained with sample data of known true correct sentences, the sample data includes at least one session data of a first sample body and a second sample data, by the method provided in the above embodiment, a sample motion vector, a sample slot position vector, and a sample portrait vector are obtained based on the at least one session data, a sample combination vector is obtained after processing (e.g., at least one of splicing and adding), the generative model is input based on the sample combination vector, the generative model is trained with the true correct sentences as supervisory information, so that the generative model can generate correct sentences and learn answer styles of different bodies, and in addition, the true correct sentences can be converted into different forms of expression by rule processing, the correct sentences with different expressions are used as supervision information, so that the generation model can learn expressions in various different modes; in the embodiment, the predicted second sentence is directly generated through the generation model without depending on a preset template, so that the problem that conversation recommendation cannot be realized without the template in the prior art is solved, and more optional sentences are provided for the second conversation main body.

As shown in fig. 5, based on the embodiment shown in fig. 1, step 106 may include the following steps:

step 1064, mapping the current motion information, the second slot position set, and the portrait information into vector expressions, respectively, to obtain a motion vector, a slot position vector, and a portrait vector.

The implementation process and technical effect of this step can refer to step 1061 in the above embodiments, and are not described herein again.

Step 1065, process the motion vector, slot position vector and image vector to obtain a combined vector.

The implementation process and technical effect of this step can refer to step 1062 in the above embodiment, and are not described herein again.

Step 1066, processing the combined vector based on the trained classification model, and determining at least one template sentence from the preset template set as at least one predicted second sentence in the predicted second sentence set.

The preset template set comprises a plurality of template sentences.

In this embodiment, the classification model may be a classification model (e.g., fasttext model, etc.) of any network structure, the classification model is trained on sample data of a known real correct sentence, the sample data includes at least one piece of session data of a first sample body and a second sample data, by the method provided in the above embodiment, a sample motion vector, a sample slot position vector, and a sample portrait vector may be obtained based on the at least one piece of session data, a sample combination vector is obtained after processing (e.g., at least one of splicing and adding), and the classification model is input based on the sample combination vector; taking the template corresponding to the correct sentence as a correct classification result, and taking other templates as wrong classification results to train the network model; so that the classification model can screen out a proper template under the condition of given input; and after selecting at least one proper template, filling the slot position fields required by the template to obtain at least one template sentence.

In addition to the embodiments shown in fig. 4 and 5 described above, the step 106 may further include all steps 1061-1066 of generating at least one predicted second sentence based on the generation model and the classification model, respectively, and determining the final target second sentence from all the predicted second sentences generated in the two methods.

As shown in fig. 6, based on the embodiment shown in fig. 1, step 108 may include the following steps:

step 1081, forming at least one predicted conversation based on the first sentence and at least one predicted second sentence included in the set of predicted second sentences.

Wherein each predicted session comprises a first sentence and a predicted second sentence, optionally, each predicted session is a question-and-answer pair.

Step 1082, scoring each of the at least one prediction session using a semantic matching model to obtain at least one evaluation score.

Wherein each rating score corresponds to a prediction session.

Alternatively, the semantic matching model may be a neural network model of any network structure, which implements a function of scoring a degree of matching between the first sentence and the predicted second sentence in the predicted conversation; the semantic matching model is trained on sample conversation data with label information, wherein the label information in each sample conversation data identifies whether a question (corresponding to a first sentence query) in the sample conversation data is answered by a question (corresponding to a second sentence answer), and when the question is answered, a label is 1, otherwise, the label is 0; taking a large number (sensor 1(query question), sensor 2 (answer), label1) as training set training models to determine whether the semantic matching model society (sensor 1, sensor 2) has matching relationship; alternatively, the rating score may be determined based on the output probability, with the higher the probability the higher the score.

Step 1083, determining n target second sentences as conversational recommendations of the second conversational partner based on each of the at least one rating score.

In the embodiment, the n predicted second sentences with higher (for example, the highest) evaluation scores are selected from the plurality of predicted second sentences and recommended to the first dialog body as the target second sentences, so that the n sentences which are most relevant to the current dialog and most accord with the conversation habits of the second dialog body are recommended, and the recommendation efficiency and the probability of selecting the recommended sentences are improved.

Optionally, step 1083 may include:

and sequencing at least one prediction session according to the magnitude of at least one evaluation score to obtain a session sequence.

And taking the predicted second sentences in the first n predicted conversations as target second sentences from the conversation sequence in sequence to obtain n target second sentences.

The embodiment provides an optional process for determining n predicted second sentences with higher score values, predicted sessions are sequenced according to the evaluation score values, the sequencing can be from large to small or from small to large, when the target second sentences are screened according to the score values, the target second sentences can be directly obtained in sequence, the n target second sentences obtained in the embodiment can be obtained from a generation model or a classification model, the problem that the session recommendation cannot be realized under the condition that no existing template exists in the prior art is solved through the generation model, the session recommendation is realized by combining two methods, the recommendation efficiency and pertinence are improved, and better session recommendation is provided for a second dialog main body.

Any of the methods for sentence recommendation in a dialog provided by embodiments of the present disclosure may be performed by any suitable device having data processing capabilities, including but not limited to: terminal equipment, a server and the like. Alternatively, the sentence recommendation method in any of the dialogs provided by the embodiments of the present disclosure may be executed by a processor, such as the processor executing the sentence recommendation method in any of the dialogs mentioned by the embodiments of the present disclosure by calling a corresponding instruction stored in a memory. And will not be described in detail below.

Exemplary devices

Fig. 7 is a schematic structural diagram of a sentence recommendation device in a dialog according to an exemplary embodiment of the present disclosure. As shown in fig. 7, the apparatus provided in this embodiment includes:

the information determining module 71 is configured to determine current intention information corresponding to a first sentence issued by a first dialog body in a current dialog, and determine current action information corresponding to a second dialog body in the current dialog based on the current intention information.

A slot determination module 72 for determining a second set of slots based on at least one historical session between the first session master and the second session master prior to the current session.

Wherein the historical conversation is a conversation between the first conversation body and the second conversation body before the current conversation; the second slot set includes at least one slot information.

An initial prediction module 73, configured to obtain a predicted second sentence set based on the current motion information, the second slot set, and the portrait information corresponding to the second main body; wherein the set of predicted second sentences includes at least one predicted second sentence.

And a conversation recommendation module 74, configured to determine n target second sentences from the predicted second sentence set as conversation recommendations of the second conversation subject.

Wherein n is an integer of 1 or more.

The sentence recommendation device in a dialog provided by the above embodiment of the present disclosure determines current intention information corresponding to a first sentence sent by a first dialog subject in a current dialog, and determines current action information corresponding to a second dialog subject in the current dialog based on the current intention information; determining a second set of slots based on at least one historical session between the first session subject and the second session subject prior to the current session; obtaining a predicted second sentence set based on the current action information, the second slot position set and the portrait information corresponding to the second main body; determining n target second sentences from the set of predicted second sentences as conversation recommendations for the second conversation subject; wherein n is an integer of 1 or more. In the embodiment, the second slot position set is determined by combining the historical conversation, so that the information in the conversation context is combined into the current conversation, the obtained target second sentence is more consistent with the current scene, the dialogue style of the second main body is combined by combining the portrait information of the second main body, personalized sentence recommendation is realized, the probability of selecting the recommended target second sentence is improved, and the conversation efficiency is further improved.

Optionally, the information determining module 71 is specifically configured to receive a first sentence, which is sent by the first dialog body in the current dialog; performing intention recognition on the first sentence by using the first network model to obtain current intention information corresponding to the first sentence; and determining the predicted action information corresponding to the current intention information by using the second network model, and taking the predicted action information as the current action information corresponding to the second dialogue main body.

Optionally, the slot determining module 72 includes:

the history conversation unit is used for acquiring at least one history first sentence and at least one history second sentence which are included in at least one history conversation; each history session comprises a session body transfer, wherein the session body comprises a first session body and a second session body;

the slot position extraction unit is used for determining a first slot position set corresponding to at least one historical first sentence and at least one historical second sentence by utilizing a third network model; wherein the first slot position set comprises at least one slot position information;

and the slot position aggregation unit is used for determining at least one slot position information from the first slot position aggregation to form a second slot position aggregation based on the current intention information and the current action information.

Optionally, the slot aggregation unit is specifically configured to determine, by using a fourth network model, a predicted slot aggregation corresponding to the current session based on the current intention information and the current action information; the prediction slot position set comprises at least one slot position information; a second set of slots is determined based on an intersection of the first set of slots and the set of predicted slots.

In some optional embodiments, the initial prediction module 73 is specifically configured to map the current motion information, the second slot position set, and the portrait information into vector expressions respectively, so as to obtain a motion vector, a slot position vector, and a portrait vector; processing the motion vector, the slot position vector and the image vector to obtain a combined vector; and processing the combined vector based on the trained generating model to obtain at least one predicted second sentence in the predicted second sentence set.

In other optional embodiments, the initial prediction module 73 is specifically configured to map the current motion information, the second slot position set, and the portrait information into vector expressions respectively, so as to obtain a motion vector, a slot position vector, and a portrait vector; processing the motion vector, the slot position vector and the image vector to obtain a combined vector; processing the combined vector based on the trained classification model, and determining at least one template sentence from a preset template set as at least one predicted second sentence in a predicted second sentence set; the preset template set comprises a plurality of template sentences.

In still other alternative embodiments, the initial prediction module 73 is specifically configured to map the current motion information, the second slot position set, and the image information into vector expressions respectively, so as to obtain a motion vector, a slot position vector, and an image vector; processing the motion vector, the slot position vector and the image vector to obtain a combined vector; processing the combined vector based on the trained generative model to obtain a predicted second sentence set comprising at least one predicted second sentence; the combined vector is processed based on the trained classification model, and at least one template sentence is determined from a preset template set to be used as at least one predicted second sentence in the predicted second sentence set; the preset template set comprises a plurality of template sentences.

Optionally, the session recommendation module 74 includes:

a prediction conversation unit configured to constitute at least one prediction conversation based on the first sentence and at least one predicted second sentence included in the predicted second sentence set;

the scoring unit is used for scoring each prediction session in the at least one prediction session by using the semantic matching model to obtain at least one evaluation score; wherein each evaluation score corresponds to a prediction session;

a target determination unit for determining n target second sentences as the conversation recommendation of the second conversation subject based on each of the at least one rating score.

Optionally, the target determining unit is specifically configured to sort the at least one predicted session according to a size of the at least one evaluation score, so as to obtain a session sequence; taking the predicted second sentences in the first n predicted conversations from the conversation sequence as target second sentences to obtain n target second sentences; and taking the n target second sentences as the conversation recommendation of the second conversation body.

Exemplary electronic device

Next, an electronic apparatus according to an embodiment of the present disclosure is described with reference to fig. 8. The electronic device may be either or both of the first device 100 and the second device 200, or a stand-alone device separate from them that may communicate with the first device and the second device to receive the collected input signals therefrom.

FIG. 8 illustrates a block diagram of an electronic device in accordance with an embodiment of the disclosure.

As shown in fig. 8, the electronic device 80 includes one or more processors 81 and memory 82.

The processor 81 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 80 to perform desired functions.

Memory 82 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by the processor 81 to implement the sentence recommendation method in dialog of the various embodiments of the present disclosure described above and/or other desired functionality. Various contents such as an input signal, a signal component, a noise component, etc. may also be stored in the computer-readable storage medium.

In one example, the electronic device 80 may further include: an input device 83 and an output device 84, which are interconnected by a bus system and/or other form of connection mechanism (not shown).

For example, when the electronic device is the first device 100 or the second device 200, the input device 83 may be a microphone or a microphone array as described above for capturing an input signal of a sound source. When the electronic device is a stand-alone device, the input means 83 may be a communication network connector for receiving the acquired input signals from the first device 100 and the second device 200.

The input device 83 may include, for example, a keyboard, a mouse, and the like.

The output device 84 may output various information including the determined distance information, direction information, and the like to the outside. The output devices 84 may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.

Of course, for simplicity, only some of the components of the electronic device 80 relevant to the present disclosure are shown in fig. 8, omitting components such as buses, input/output interfaces, and the like. In addition, the electronic device 80 may include any other suitable components depending on the particular application.

Exemplary computer program product and computer-readable storage Medium

In addition to the above-described methods and apparatus, embodiments of the present disclosure may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the sentence recommendation method in dialog according to various embodiments of the present disclosure described in the "exemplary methods" section of this specification above.

The computer program product may write program code for carrying out operations for embodiments of the present disclosure in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

Furthermore, embodiments of the present disclosure may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform steps in a sentence recommendation method in a dialog according to various embodiments of the present disclosure described in the "exemplary methods" section above of this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing describes the general principles of the present disclosure in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present disclosure are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present disclosure. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the disclosure is not intended to be limited to the specific details so described.

In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts in the embodiments are referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The block diagrams of devices, apparatuses, systems referred to in this disclosure are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

The methods and apparatus of the present disclosure may be implemented in a number of ways. For example, the methods and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

It is also noted that in the devices, apparatuses, and methods of the present disclosure, each component or step can be decomposed and/or recombined. These decompositions and/or recombinations are to be considered equivalents of the present disclosure.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit embodiments of the disclosure to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. A method for sentence recommendation in a dialog, comprising:

2. The method of claim 1, wherein determining current intention information corresponding to a first sentence uttered by a first dialog body in a current dialog, and determining current action information corresponding to a second dialog body in the current dialog based on the current intention information comprises:

3. The method of claim 1 or 2, wherein determining a second set of slots based on at least one historical session between the first session master and the second session master prior to the current session comprises:

4. The method of claim 3, wherein the determining at least one slot information from the first set of slots to constitute a second set of slots based on the current intent information and the current action information comprises:

5. The method of any of claims 1-4, wherein obtaining a predicted second set of sentences based on the current action information, the second set of slots, and the portrait information corresponding to the second subject comprises:

6. The method of any of claims 1-5, wherein obtaining the predicted second set of sentences based on the current action information, the second set of slots, and the portrait information corresponding to the second subject comprises:

7. The method according to any of claims 1-6, wherein said determining n target second sentences from said set of predicted second sentences as conversational recommendations for said second conversational partner comprises:

8. The method of claim 7, wherein the determining the n target second sentences as the conversational recommendation of the second conversational partner based on each of the at least one opinion score comprises:

9. A computer-readable storage medium, characterized in that the storage medium stores a computer program for executing the sentence recommendation method in a dialog according to any one of the above claims 1-8.

10. An electronic device, characterized in that the electronic device comprises:

a processor;

a memory for storing the processor-executable instructions;

the processor is configured to read the executable instructions from the memory and execute the instructions to implement the sentence recommendation method in dialog of any of claims 1-8.