CN116702736A

CN116702736A - Safe call generation method and device, electronic equipment and storage medium

Info

Publication number: CN116702736A
Application number: CN202310828238.2A
Authority: CN
Inventors: 吴岸城
Original assignee: Ping An Life Insurance Company of China Ltd
Current assignee: Ping An Life Insurance Company of China Ltd
Priority date: 2023-07-07
Filing date: 2023-07-07
Publication date: 2023-09-05

Abstract

The embodiment of the application provides an insurance policy generation method, an insurance policy generation device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring object basic information and risk focus information of a target object; the risk type attention information is used for representing attention information of the target object to a preset risk type; performing feature coding on the object basic information to obtain object features; screening out a selected speaking template text from a preset insurance template database according to the risk type attention information; the selected speaking template text is used for representing the universal speaking of the preset dangerous seed; performing feature coding on the text of the selected conversation template to obtain safety conversation template features; inputting the object characteristics and the safe conversation template characteristics into a preset conversation generation model for conversation generation to obtain a target safe conversation. The embodiment of the application can provide customized insurance techniques for different objects.

Description

Safe call generation method and device, electronic equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence technologies, and in particular, to a method and apparatus for generating an insurance policy, an electronic device, and a storage medium.

Background

In the process of insurance sales, sales personnel need to guide clients to complete the flow of insurance sales, introduce themselves to clients, analyze demands, fully explain products, promote sales and the like.

In the existing insurance system, how to explain the product to promote sales can be summarized as a unified conversation which plays an important role in the promotion of daily user communication. The current speech induction summary work often requires manual writing of required sales speech according to the characteristics and amount of insurance products and the like. However, these sales techniques are popular, and often cannot be used for product recommendation according to the actual demands of customers, so that the sales techniques cannot be used for bill making or delay in bill making. Therefore, how to provide an insurance policy generation method, which can provide customized insurance policies for different objects, is a technical problem to be solved.

Disclosure of Invention

The embodiment of the application mainly aims to provide an insurance policy generation method, an insurance policy generation device, electronic equipment and a storage medium, which can provide customized insurance policies for different objects.

To achieve the above object, a first aspect of an embodiment of the present application provides an insurance policy generation method, including:

Acquiring object basic information and risk focus information of a target object; the risk type attention information is used for representing attention information of the target object to a preset risk type;

performing feature coding on the object basic information to obtain object features;

screening out a selected speaking template text from a preset insurance template database according to the risk type attention information; the selected speaking template text is used for representing the universal speaking of the preset dangerous seed;

performing feature coding on the text of the selected conversation template to obtain safety conversation template features;

inputting the object characteristics and the safe conversation template characteristics into a preset conversation generation model for conversation generation to obtain a target safe conversation.

In some embodiments, inputting the object feature and the safe call template feature into a preset call generation model for call generation to obtain a target safe call, including:

vector encoding is carried out on the object features to obtain object word sequence vectors;

vector coding is carried out on the safety speaking template features to obtain a speaking template word sequence vector;

performing context prediction according to the object word sequence vector and the word sequence vector of the word operation template to obtain a predicted word sequence vector;

Decoding the predicted speech word sequence vector to obtain the target insurance speech; the target safe speaking is used for representing the non-universal speaking of the preset dangerous seed.

In some embodiments, the object word sequence vector includes at least one object word vector, the speech template word sequence vector includes at least one candidate speech template word vector, and the context prediction is performed according to the object word sequence vector and the speech template word sequence vector, to obtain a predicted speech word sequence vector, including:

randomly covering the word vectors of the candidate conversation template to obtain word vectors of the selected conversation template;

acquiring the position of the selected speech template word vector in the speech template word sequence vector to obtain a selected position;

filling the object word vector to the selected position so as to update the word sequence vector of the speech operation template to obtain an intermediate word sequence vector;

sentence forming discrimination is carried out on the intermediate word sequence vector to obtain discrimination results;

and if the judging result is that the intermediate word sequence vector can form sentences, obtaining the predicted word sequence vector according to the intermediate word sequence vector.

In some embodiments, before inputting the object feature and the safe call template feature into a preset call generation model to perform call generation, and obtaining a target safe call, the method further includes:

training the speech generation model, comprising:

constructing a sample set; the samples in the sample set comprise a first sample safe and a second sample safe, the first sample safe and the second sample safe are different from the same preset dangerous, and the number of words of the first sample safe is smaller than that of the second sample safe;

performing feature coding on the first sample safe call to obtain the features of the sample call to be processed;

vector conversion is carried out on the second sample safe operation, and a reference sample safe operation feature vector is obtained;

performing voice operation generation on the to-be-processed sample voice operation characteristics through a preset initial generation model to obtain a predicted sample safety voice operation;

vector conversion is carried out on the prediction sample safe call operation, and a prediction sample safe call operation feature vector is obtained;

performing loss calculation according to the predicted sample microphone feature vector and the reference sample microphone feature vector to obtain loss data;

And carrying out parameter adjustment on the initial generation model according to the loss data to obtain the speaking operation generation model.

In some embodiments, before the screening of the selected conversation template text from the pre-set insurance template database according to the risk category interest information, the method further includes:

constructing the insurance template database, including:

acquiring insurance product data of the preset dangerous seed; wherein the insurance product data includes at least one candidate text segment;

text screening is carried out on the candidate text segments according to preset text screening conditions, and selected text segments are obtained; wherein the selected text segment includes at least one candidate information sentence;

sentence screening is carried out on the candidate information sentences according to preset sentence screening conditions, and selected information sentences are obtained;

extracting keywords from the selected information sentences to obtain at least one selected keyword;

filling the selected keywords into a preset information extraction template to obtain a candidate conversation template text;

and merging the candidate speech operation template texts to obtain the insurance template database.

In some embodiments, the sentence screening for the candidate information sentences according to a preset sentence screening condition, to obtain a selected information sentence, includes:

Acquiring the attention score of each candidate information sentence through a preset attention model;

and if the attention score is larger than a preset attention score threshold, obtaining the selected information sentence according to the candidate information sentence.

In some embodiments, before filling the selected keywords into a preset information extraction template to obtain the candidate speech template text, the method further includes:

filtering the selected keywords, including:

selecting one from at least two selected keywords to obtain a first keyword, and selecting the other from at least two selected keywords to obtain a second keyword;

performing similarity calculation on the first keyword and the second keyword to obtain reference similarity;

and if the reference similarity is larger than a preset similarity threshold, deleting the first keyword or the second keyword to obtain the filtered selected keyword.

To achieve the above object, a second aspect of an embodiment of the present application provides an insurance session generating device, including:

the information acquisition module is used for acquiring object basic information and risk type attention information of the target object; the risk type attention information is used for representing attention information of the target object to a preset risk type;

The object feature coding module is used for carrying out feature coding on the object basic information to obtain object features;

the conversation template searching module is used for screening out a selected conversation template text from a preset insurance template database according to the risk attention information; the selected speaking template text is used for representing the universal speaking of the preset dangerous seed;

the template feature coding module is used for carrying out feature coding on the selected voice template text to obtain safety voice template features;

and the conversation generation module is used for inputting the object characteristics and the safety conversation template characteristics into a preset conversation generation model to perform conversation generation so as to obtain a target safety conversation.

To achieve the above object, a third aspect of the embodiments of the present application provides an electronic device, which includes a memory and a processor, where the memory stores a computer program, and the processor implements the method for generating an insurance policy according to the first aspect when executing the computer program.

To achieve the above object, a fourth aspect of the embodiments of the present application proposes a storage medium, which is a computer-readable storage medium storing a computer program that when executed by a processor implements the method for generating an safe call according to the first aspect.

The application provides an insurance policy generation method, an insurance policy generation device, electronic equipment and a storage medium, wherein the method comprises the following steps: searching the insurance template database through the risk type attention information to obtain a selected conversation template text, wherein the selected conversation template text is used for representing a universal conversation of a preset risk type. The object basic information is encoded to obtain object characteristics, and the selected conversation template text is encoded to obtain insurance conversation template characteristics. The speech generation model of the embodiment of the application performs speech generation according to the object characteristics and the safe speech template characteristics, which is equivalent to speech expansion by utilizing the object characteristics and the safe speech template characteristics. In summary, the embodiment of the application can generate different target safe dialogs based on different object characteristics, and can provide customized safe dialogs for different objects.

Drawings

FIG. 1 is a schematic diagram of a system architecture for performing an insurance policy generation method according to an embodiment of the present application;

FIG. 2 is a flow chart of an insurance policy generation method provided by an embodiment of the present application;

FIG. 3 is a flow chart of an insurance session generating method according to another embodiment of the present application;

FIG. 4 is a flow chart of an insurance session generating method according to another embodiment of the present application;

FIG. 5 is a flow chart of an insurance session generating method according to another embodiment of the present application;

fig. 6 is a flowchart of step S105 in fig. 2;

fig. 7 is a flowchart of step S503 in fig. 6;

FIG. 8 is a block diagram of a module structure of an apparatus for generating an insurance session according to an embodiment of the present application;

fig. 9 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the present application.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

It should be noted that although functional block division is performed in a device diagram and a logic sequence is shown in a flowchart, in some cases, the steps shown or described may be performed in a different order than the block division in the device, or in the flowchart. The terms first, second and the like in the description and in the claims and in the above-described figures, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the application only and is not intended to be limiting of the application.

First, several nouns involved in the present application are parsed:

artificial intelligence (artificial intelligence, AI): is a new technical science for researching and developing theories, methods, technologies and application systems for simulating, extending and expanding the intelligence of people; artificial intelligence is a branch of computer science that attempts to understand the nature of intelligence and to produce a new intelligent machine that can react in a manner similar to human intelligence, research in this field including robotics, language recognition, image recognition, natural language processing, and expert systems. Artificial intelligence can simulate the information process of consciousness and thinking of people. Artificial intelligence is also a theory, method, technique, and application system that utilizes a digital computer or digital computer-controlled machine to simulate, extend, and expand human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results.

Natural language processing (natural language processing, NLP): NLP is a branch of artificial intelligence that is a interdisciplinary of computer science and linguistics, and is often referred to as computational linguistics, and is processed, understood, and applied to human languages (e.g., chinese, english, etc.). Natural language processing includes parsing, semantic analysis, chapter understanding, and the like. Natural language processing is commonly used in the technical fields of machine translation, handwriting and print character recognition, voice recognition and text-to-speech conversion, information image processing, information extraction and filtering, text classification and clustering, public opinion analysis, and viewpoint mining, and the like, and relates to data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, linguistic research related to language calculation, and the like.

In the process of insurance sales, sales personnel need to guide clients to complete the flow of insurance sales, introduce themselves to clients, analyze demands, fully explain products, promote sales and the like. In the existing insurance system, how to explain the product to promote sales can be summarized as a unified conversation which plays an important role in the promotion of daily user communication. At present, the induction and summarization work of the speaking operation often needs to manually write the required product explanation and the speaking operation of sales according to the characteristics, the amount and the like of insurance products. However, the sales of manually written products is related to personal knowledge structures and understanding of insurance products, and it often happens that the writing of the sales does not match the customers of the corresponding personality or age level, so that the sales cannot be made into a bill or is delayed to be made into a bill. In addition, there are new insurance products each month, and manual management is not an easy task for all insurance products, nor is anyone able to do. Finally, the agent needs to generate a proper conversation according to the characteristics of the user. Therefore, how to provide an insurance policy generation method, which can provide customized insurance policies for different objects, is a technical problem to be solved.

Based on the above, the embodiment of the application provides an insurance speech generating method, an insurance speech generating device, electronic equipment and a storage medium, which aim to perform speech generation on object features and speech template features through a speech generating model, can provide customized insurance speech for different objects, improve the use experience of users and increase the insurance policy.

The method for generating the safe call provided by the embodiment of the application is applied to the server side and can also be software running in the server side. The server side can be configured as an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligent platforms and the like; the software may be an application or the like that implements the method of creation of the safe call, but is not limited to the above form.

The application is operational with numerous general purpose or special purpose computer system environments or configurations. For example: server computers, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, distributed computing environments that include any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiment of the application provides an insurance policy generation method, an insurance policy generation device, electronic equipment and a storage medium, and specifically, the following embodiment is used for explaining, and first describes the insurance policy generation method in the embodiment of the application.

Referring to fig. 1, the method for generating the safe call according to the embodiment of the present application may be performed by the target server 400 alone, may be performed by the first terminal 100, the second terminal 200, or the third terminal 300 alone, or may be performed by the first terminal 100, the second terminal 200, the third terminal 300, and the target server 400 together.

In the embodiments of the present application, when related processing is performed on data related to the identity or characteristics of a user according to basic information of the object, permission or consent of the user is obtained first, and the collection, use and processing of the data comply with related laws and regulations and standards.

Fig. 2 is an alternative flowchart of an embodiment of the method for generating an insurance session according to the present application, which may include, but is not limited to, steps S101 to S105.

Step S101, object basic information and risk attention information of a target object are obtained; the risk type attention information is used for representing attention information of a target object to a preset risk type;

Step S102, performing feature coding on object basic information to obtain object features;

step S103, screening out a selected speaking template text from a preset insurance template database according to risk type attention information; the selected speaking template text is used for representing a universal speaking of a preset dangerous seed;

step S104, performing feature coding on the text of the selected voice operation template to obtain features of the safety voice operation template;

step S105, inputting the object characteristics and the safe conversation template characteristics into a preset conversation generation model for conversation generation to obtain a target safe conversation.

In the steps S101 to S105 shown in the embodiment of the application, the insurance template database is searched through the risk attention information to obtain the text of the selected speaking template. The selected conversation template text is used for representing the general conversation of the preset dangerous types and generally comprises general description text such as product names, responsibility ranges and the like, but does not comprise non-general description text related to object basic information. The object basic information is encoded to obtain object characteristics, and the selected conversation template text is encoded to obtain insurance conversation template characteristics. The speech generation model of the embodiment of the application performs speech generation according to the object characteristics and the safety speech template characteristics, which is equivalent to speech expansion by utilizing the object characteristics and the safety speech template characteristics, and the obtained target safety speech comprises a non-universal speech generated based on the object characteristics. In summary, the embodiment of the application can generate different target insurance policies based on different object characteristics, so that customized insurance policies can be provided for different objects.

In step S101 of some embodiments, object base information and risk focus information of a target object are acquired; the risk type attention information is used for representing attention information of the target object to a preset risk type.

Specifically, the target object may be a crowd who has been consulted or underway for insurance service at an insurance website or related off-line website. The target object can be determined according to the insurance business records by acquiring the insurance business records of the insurance website or the off-line website system.

After the target object is determined, object base information is acquired. The object base information can be regarded as a description of the target object. For example, the object basic information includes character information, property information, and consumption information. The character information includes name, age, sex, region, education level, family composition, insurance preference, etc. The consumption data includes shopping records, banking flows, etc.

After the target object is determined, risk type attention information is acquired, wherein the risk type attention information is used for representing the attention information of the target object to a preset risk type. It should be noted that, in the case of generating an insurance session, it is necessary to determine in advance which kind of risk the target object is, so as to improve the accuracy of the insurance session. The risk care information includes insurance product name, insurance product type, etc. Different target objects and corresponding risk focus information are different. For example, if the shopping record of the object a includes a vehicle product, the object a may be considered as paying attention to the vehicle insurance product, and the risk information of the object a is the vehicle insurance name, the vehicle insurance amount and the vehicle insurance premium. Thus, the risk focus information can be acquired through the shopping record of the target object. For another example, the risk concern information may be obtained through an insurance consultation record of the target object.

In step S102 of some embodiments, feature encoding is performed on the object base information to obtain object features. Specifically, the object basic information is subjected to feature coding through an encoder to obtain object features.

In some embodiments, referring to fig. 3, before step S103, the method for generating an insurance policy according to the embodiment of the present application further includes: constructing an insurance template database, including in particular but not limited to steps S201 to S206:

step S201, acquiring insurance product data of a preset dangerous seed; wherein the insurance product data includes at least one candidate text segment;

step S202, text screening is carried out on candidate text segments according to preset text screening conditions, and selected text segments are obtained; wherein the selected text segment includes at least one candidate information sentence;

step S203, sentence screening is carried out on candidate information sentences according to preset sentence screening conditions to obtain selected information sentences;

step S204, extracting keywords from the selected information sentences to obtain at least one selected keyword;

step S205, filling the selected keywords into a preset information extraction template to obtain a candidate speech operation template text;

and S206, merging the candidate speech operation template texts to obtain an insurance template database.

In step S201 of some embodiments, insurance product data of a preset risk is extracted from a product database, and the insurance product data may be a product information table, a product clause, or a product specification. It should be noted that, when the insurance product data is obtained, a simple table and punctuation mark cleaning can be performed.

In step S202 of some embodiments, text filtering is performed on candidate text segments according to preset text filtering conditions, so as to obtain selected text segments. Specifically, the text filtering conditions include: if the candidate text segment contains key information, it is taken as the selected text segment. The key information refers to product introduction, insurance responsibility, responsibility relief, cost, rate and the like. And carrying out sentence splitting on the selected text segment according to the punctuation to obtain at least one candidate information sentence.

In step S203 of some embodiments, sentence screening is performed on candidate information sentences according to preset sentence screening conditions, so as to obtain selected information sentences. Sentence screening conditions include: key information. If the candidate information sentence includes key information, it is regarded as the selected information sentence. In one example, a classification model may be pre-trained, candidate information sentences are classified by the classification model, and selected information sentences are determined based on the classification result.

In some embodiments, the sentence screening conditions include an attention score, and step S203 specifically includes, but is not limited to:

The selected information sentences are selected based on the attention score, the sentences with high attention score are high in importance, and the expressed speech characteristics are more representative of the real situation of insurance products than other sentences, so that the accuracy of insurance speech generation is improved.

Specifically, the attention model refers to a model that when a certain problem is solved in a specific scene, different attention scores are applied to different information which needs to be considered for solving the problem, higher weight is applied to information with large help of the problem, lower weight is applied to information with small help of the problem, and therefore the problem is solved by using the information better. In particular, for embodiments of the present application, there are multiple candidate information sentences in a selected text passage, each of which contributes significantly to the insured speech generation of such a question. For the safe-guard generation of sentences that contribute significantly, a greater attention score is given by the attention model. For safe speech generation sentences that contribute little are given a smaller attention score by the attention model.

The attention model is generally input by taking a binary sequence as input, and each candidate information sentence of the selected text segment is text, so that the candidate information sentence is segmented by utilizing a word segmentation technology, the segmented words are converted into word vectors, and the word vectors are spliced according to the sequence of the words in the sentence, so that the sentence vectors are obtained. The word is converted into a word vector by a one-hot (one-hot) vector representation method and a word embedding (word embedding) method.

In step S204 of some embodiments, keyword extraction is performed on the selected information sentence to obtain at least one selected keyword. For example, the selected information sentence is compared with a preset keyword list, and the key short sentences and phrases in the selected information sentence, such as texts containing the key information including the amount of money, the insurance range, the underwriting age, the notice and the like, are taken out to obtain the selected keyword.

In some embodiments, referring to fig. 4, before step S205, the method for generating an insurance policy according to the embodiment of the present application further includes:

filtering the selected keywords, including but not limited to steps S301 to S303:

step S301, selecting one from at least two selected keywords to obtain a first keyword, and selecting the other from the at least two selected keywords to obtain a second keyword;

Step S302, similarity calculation is carried out on the first keyword and the second keyword, and reference similarity is obtained;

step S303, deleting the first keyword or the second keyword if the reference similarity is larger than a preset similarity threshold value, and obtaining the filtered selected keyword.

In the steps S301 to S303 shown in the embodiment of the present application, mainly selected keywords with high similarity are combined, and the selected keywords with high variability are remained as much as possible. Therefore, when the reference similarity is larger than the similarity threshold, deleting the first keyword or the second keyword, so that only one of the first keyword and the second keyword with higher similarity can be left.

In step S205 of some embodiments, the selected keywords are filled into a preset information extraction template, resulting in a candidate speech template text. For example, according to the information extraction template, extracting information such as specific amount, specific insurance range, specific insurance age and the like in the selected keywords; and reserving information such as placeholders according to the conversation characteristics, and fusing the key information extracted in the previous step to obtain the candidate conversation template text.

In one example, the information extraction template is "[ subject name ], which is [ insurance category ], the insurance scope used [ insurance scope ], there are additional rates per year [ additional rates ]. The selected keywords include: insurance product M, serious illness, 6-60 years old, 3%. The obtained candidate speech template text is "[ object name ], the insurance product M is a serious insurance, the used insurance range is 6-60 years old, and the additional rate is 3% per year. It should be noted that, the candidate phone template text in the embodiment of the present application is used to represent the universal phone of each insurance product, and does not include object basic information, for example, does not include an object name, and cannot embody the customized phone of different objects.

In step S206 of some embodiments, the candidate speech template text is merged to obtain an insurance template database. Specifically, the text of the candidate speech operation template is stored in a preset initial database to obtain an insurance template database.

Through steps S201 to S206 illustrated in the embodiment of the present application, the insurance template database has been successfully created. In the embodiment of the application, the speaking operation generation is divided into two parts, and the template generation and the speaking operation generation are performed, so that the object basic information of the target object and the insurance product basic speaking operation can be distinguished, and the speaking operation generation model is utilized to generate the non-universal speaking operation aiming at the target object to the greatest extent.

In step S103 of some embodiments, selecting a selected conversation template text from a preset insurance template database according to risk focus information; wherein, the selected speaking template text is used for representing the universal speaking of the preset dangerous seed.

Specifically, the insurance template database stores a plurality of candidate phone template texts, each candidate phone template text is used for representing a general phone of an insurance product, and the candidate phone template texts generally comprise general phones such as product names, product premium, product insurance amounts, product insurance ranges, product premium payment modes and the like. Taking the preset dangerous seed as an example of the vehicle insurance, if the dangerous seed attention information is the vehicle insurance name, the vehicle insurance amount or the vehicle insurance premium, screening the candidate phone-call template text according to the vehicle insurance name, the vehicle insurance amount or the vehicle insurance premium, and obtaining the selected phone-call template text with the preset dangerous seed as the vehicle insurance. The text of the selected conversation template can be "[ object name ]", and the insurance product X is a car insurance, a car insurance amount Y and a car insurance premium Z ".

In an example, the insurance interest information includes a product name, and step S103 specifically includes: comparing the insurance name with the candidate speech operation template text of the insurance template database to obtain a comparison result; and if the comparison result is that the insurance name exists in the candidate phone template text, the candidate phone template text is used as the selected phone template text.

In step S104 of some embodiments, feature encoding is performed on the selected voice template text to obtain safe voice template features. Specifically, the selected speech template text is subjected to feature coding through an encoder to obtain the features of the safe speech template.

In some embodiments, referring to fig. 5, before step S105, the method for generating an insurance policy according to the embodiment of the present application further includes:

training session generation models, including but not limited to steps S401 through S407:

step S401, a sample set is constructed; the samples in the sample set comprise a first sample safe and a second sample safe, the first sample safe and the second sample safe are different from the same preset dangerous seed, and the number of words of the first sample safe is smaller than that of the second sample safe;

Step S402, performing feature coding on a first sample safe call to obtain features of the sample call to be processed;

step S403, carrying out vector conversion on the second sample safe operation to obtain a reference sample safe operation feature vector;

step S404, performing voice operation generation on the characteristics of the sample voice operation to be processed through a preset initial generation model to obtain a predicted sample safety voice operation;

step S405, vector conversion is carried out on the predicted sample safe operation to obtain a predicted sample safe operation feature vector;

step S406, performing loss calculation according to the predicted sample speech feature vector and the reference sample speech feature vector to obtain loss data;

and step S407, carrying out parameter adjustment on the initial generation model according to the loss data to obtain a speaking operation generation model.

Through steps S401 to S407 illustrated in the embodiment of the present application, a first sample safe call is used as a basis, and a call is generated to obtain a predicted sample safe call. And the second sample safe call is taken as a reference, and parameters of the initial generation model are adjusted, so that the predicted sample safe call approaches the second sample safe call, and the call generation model is finally obtained.

It should be noted that the initial generation model may be a unified language model (Unified Language Model Pre-training for Natural Language Understanding and Generation, unilm). The speech generation model may also be a text generation model (Bidirectional Encoder Representation from Transformers, bert). It should be noted that, compared with the Bert model, the Unilm model combines the features of the self-coding model and the autoregressive model, so that the generated target safe speaking is more smooth.

In step S401 of some embodiments, the insurance database stores various kinds of utterances related to the preset dangerous seed, and thus, the first sample and second sample utterances of the preset dangerous seed may be acquired from the insurance database. It should be noted that both the first sample safe and the second sample safe include a general introduction text to the preset dangerous category, but the second sample safe may include more information, so the number of words of the second sample safe is greater than that of the first sample safe. For example, the first sample safe call may be similar to the candidate call template text, with only a generic introduction to the product, but no non-generic call related to the underlying information of the object. While a second sample safe may be similar to the target safe of embodiments of the present application, including predictive sessions generated based on object characteristics, in addition to general introduction of products.

In step S406 of some embodiments, a loss calculation is performed according to the predicted sample microphone feature vector and the reference sample microphone feature vector, so as to obtain loss data. For example, the euclidean distance between the predicted and reference sample microphone feature vectors may be calculated, and data is indeed lost based on the euclidean distance. The cosine distance may also be calculated, from which the loss data is determined.

In step S105 of some embodiments, the object feature and the safe call template feature are input to a preset call generation model for call generation, so as to obtain a target safe call.

In one example, for a target object, it may be known from a purchase record whether to own a home car. If the household vehicle is owned, the target object can be considered to pay attention to the vehicle risk. And for the target object, acquiring object basic information, and considering that dangerous information of the target object on the vehicle insurance product is in vehicle insurance premium. And screening the text of the candidate speech operation template according to the insurance premium. A selected talk pattern text is obtained that includes a generic introduction to the insurance premium and the product warranty for the vehicle insurance. Of course, the selected conversation template text also comprises some blank places, the blank places are subjected to context prediction according to the object characteristics and the selected conversation template characteristics in the subsequent conversation generation to obtain predicted words, and the blank places are filled according to the predicted words to obtain the target safe conversation in the embodiment of the application.

In some embodiments, referring to fig. 6, step S105 specifically includes, but is not limited to, steps S501 to S504:

Step S501, carrying out vector coding on object features to obtain object word sequence vectors;

step S502, vector coding is carried out on the features of the insurance speech template to obtain speech template word sequence vectors;

step S503, performing context prediction according to the object word sequence vector and the speech operation template word sequence vector to obtain a predicted speech operation word sequence vector;

step S504, decoding the predicted speech word sequence vector to obtain a target insurance speech; target insurance policies are used to represent non-generic policies for pre-set risk.

The speech generating model specifically includes a processing layer, a prediction layer and a decoding layer which are sequentially connected in steps S501 to S504 shown in the embodiment of the present application. The processing layer may receive the object feature and the safe speech template feature, and the output result includes the object word sequence vector and the speech template word sequence vector. The prediction layer receives the target word sequence vector and the word sequence vector of the word operation template, and the output result is the predicted word sequence vector. The decoding layer receives the predicted word sequence vector and outputs the result which is the target insurance policy.

In one example, the acquired object base information includes family composition information including parents, children. The acquired risk information is disease risk. Assuming that the text of the selected conversation template obtained according to the risk attention information is "[ object name ], the insurance product M is a serious disease insurance, the used insurance range is 6-60 years old, and the additional rate is 3% per year. After the above steps S501 to S504, the target insurance session may be "mr. He, the insurance product M is a serious accident, the insurance range used is 6-60 years old, both parents and children can purchase, and additional rate is 3% per year".

In some embodiments, the object word sequence vector includes at least one object word vector, the speech template word sequence vector includes at least one candidate speech template word vector, referring to fig. 7, step S503 includes, but is not limited to, steps S601 to S605:

step S601, randomly covering word vectors of the candidate speech operation templates to obtain word vectors of the selected speech operation templates;

step S602, obtaining the position of the selected speaking template word vector in the speaking template word sequence vector to obtain a selected position;

step S603, filling the object word vector to a selected position, and updating the object word vector by using a dialogue template word sequence vector to obtain an intermediate word sequence vector;

step S604, sentence-forming discrimination is carried out on the intermediate word sequence vector to obtain discrimination results;

in step S605, if the discrimination result is that the intermediate word sequence vector can be sentence-formed, the predicted word sequence vector is obtained according to the intermediate word sequence vector.

In the steps S601 to S605 shown in the embodiment of the present application, the speech template word sequence vector includes a plurality of candidate speech template word vectors spliced in sequence, the candidate speech template word vectors are randomly covered, and the covered candidate speech template word vectors are used as selected speech template word vectors. It should be noted that the number of the substrates, the masking method can be used for each selected speech template word vector the context of the word sequence vector of the speech template is seen. And then, the position of the selected speaking template word vector in the speaking template word sequence vector is obtained, and the selected position is obtained. And then filling the object word vector into the selected position, which is equivalent to replacing the selected speech operation template word vector with the object word vector, thereby obtaining the intermediate word sequence vector. And then sentence forming judgment is carried out on the intermediate word sequence vector, and if the judgment result is that the intermediate word sequence vector can form sentences, a predicted word sequence vector is obtained according to the intermediate word sequence vector. It can be appreciated that, in the embodiment of the present application, the intermediate word sequence vector can be obtained by means of random masking and word vector replacement, and the intermediate word sequence vector can be actually directly used as the predicted word sequence vector. However, in order to ensure the smoothness of sentences, the application needs to perform sentence-forming judgment on the intermediate word sequence vector so as to generate more smooth target insurance techniques.

In one embodiment, sentence-forming discrimination can be performed on the intermediate word sequence vector through the Bert model to obtain a discrimination result. The discrimination result includes that the intermediate word sequence vector can form sentences, and also includes that the intermediate word sequence vector cannot form sentences. If the discrimination result is that the intermediate word sequence vector cannot be sentence-formed, the selected word operation template word vector can be redetermined so as to redetermine the intermediate word sequence vector. If the discrimination result is that the intermediate word sequence vector can not be sentence-formed, the object word vector can be replaced so as to redetermine the intermediate word sequence vector.

Referring to fig. 8, an embodiment of the present application further provides an insurance policy generation device, which may implement the foregoing insurance policy generation method, and fig. 8 is a block diagram of a module structure of the insurance policy generation device provided by the embodiment of the present application, where the device includes: an information acquisition module 701, an object feature encoding module 702, a speech template searching module 703, a template feature encoding module 704 and a speech generating module 705. The information acquisition module 701 is configured to acquire object basic information and risk attention information of a target object; the risk type attention information is used for representing attention information of a target object to a preset risk type; the object feature encoding module 702 is configured to perform feature encoding on the object basic information to obtain object features; the speaking template searching module 703 is used for screening out a text of the selected speaking template from a preset insurance template database according to the risk type attention information; the selected speaking template text is used for representing a universal speaking of a preset dangerous seed; the template feature encoding module 704 is configured to perform feature encoding on the text of the selected conversation template to obtain features of the safe conversation template; the speech generation module 705 is configured to input the object feature and the safe speech template feature to a preset speech generation model for speech generation, so as to obtain a target safe speech.

The safe speaking operation generating device provided by the embodiment of the application further comprises a model training module and a database construction module, wherein the model training module is used for training a speaking operation generating model. The database construction module is used for constructing an insurance template database.

It should be noted that, the specific implementation of the safe call generation device is basically the same as the specific embodiment of the safe call generation method, and will not be described herein.

The embodiment of the application also provides electronic equipment, which comprises: the system comprises a memory, a processor, a program stored on the memory and capable of running on the processor, and a data bus for realizing connection communication between the processor and the memory, wherein the program realizes the method for generating the safe call when being executed by the processor. The electronic equipment can be any intelligent terminal including a tablet personal computer, a vehicle-mounted computer and the like.

Referring to fig. 9, fig. 9 illustrates a hardware structure of an electronic device according to another embodiment, the electronic device includes:

the processor 801 may be implemented by a general purpose CPU (Central Processing Unit ), microprocessor, application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc. for executing relevant programs to implement the technical scheme provided by the embodiments of the present application;

The Memory 802 may be implemented in the form of a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a random access Memory (Random Access Memory, RAM). Memory 802 may store an operating system and other application programs, and when the technical solutions provided in the embodiments of the present disclosure are implemented by software or firmware, relevant program codes are stored in memory 802, and the processor 801 invokes an insurance policy generation method for executing the embodiments of the present disclosure;

an input/output interface 803 for implementing information input and output;

the communication interface 804 is configured to implement communication interaction between the device and other devices, and may implement communication in a wired manner (e.g., USB, network cable, etc.), or may implement communication in a wireless manner (e.g., mobile network, WIFI, bluetooth, etc.);

a bus 805 that transfers information between the various components of the device (e.g., the processor 801, the memory 802, the input/output interface 803, and the communication interface 804);

wherein the processor 801, the memory 802, the input/output interface 803, and the communication interface 804 implement communication connection between each other inside the device through a bus 805.

The embodiment of the application also provides a storage medium, which is a computer readable storage medium and is used for computer readable storage, the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to realize the safe call generation method.

The memory, as a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory remotely located relative to the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

According to the safe operation generating method, the safe operation generating device, the electronic equipment and the storage medium provided by the embodiment of the application, the safe template database is searched through dangerous type attention information, and the text of the selected operation template is obtained. The selected conversation template text is used for representing the general conversation of the preset dangerous types and generally comprises general description text such as product names, responsibility ranges and the like, but does not comprise non-general description text related to object basic information. The object basic information is coded to obtain object characteristics, and the selected conversation template text is coded to obtain insurance conversation template characteristics. The speech generation model of the embodiment of the application performs speech generation according to the object characteristics and the safety speech template characteristics, which is equivalent to speech expansion by utilizing the object characteristics and the safety speech template characteristics, and the obtained target safety speech comprises a non-universal speech generated based on the object characteristics. In summary, the embodiment of the application can generate different target insurance policies based on different object characteristics, so that customized insurance policies can be provided for different objects.

The embodiments described in the embodiments of the present application are for more clearly describing the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application, and those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

It will be appreciated by those skilled in the art that the solutions shown in fig. 2-7 are not limiting on the embodiments of the application and may include more or fewer steps than shown, or certain steps may be combined, or different steps.

The above described apparatus embodiments are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Those of ordinary skill in the art will appreciate that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof.

The terms "first," "second," "third," "fourth," and the like in the description of the application and in the above figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be understood that in the present application, "at least one (item)" means one or more, and "a plurality" means two or more. "and/or" for describing the association relationship of the association object, the representation may have three relationships, for example, "a and/or B" may represent: only a, only B and both a and B are present, wherein a, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (one) of a, b or c may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown 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 may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including multiple instructions for causing an electronic device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing a program.

The preferred embodiments of the present application have been described above with reference to the accompanying drawings, and are not thereby limiting the scope of the claims of the embodiments of the present application. Any modifications, equivalent substitutions and improvements made by those skilled in the art without departing from the scope and spirit of the embodiments of the present application shall fall within the scope of the claims of the embodiments of the present application.

Claims

1. A method of generating an insurance session, the method comprising:

2. The method of claim 1, wherein inputting the object feature and the safe call template feature into a preset call generation model for call generation to obtain a target safe call comprises:

3. The method of claim 2, wherein the sequence of object words vector comprises at least one sequence of object words vector, the sequence of speech templates word vector comprises at least one candidate speech templates word vector, and wherein performing context prediction based on the sequence of object words vector and the sequence of speech templates word vector to obtain a predicted sequence of speech words vector comprises:

4. The method of claim 1, wherein prior to the inputting the object features and the safe-call template features into a preset safe-call generation model for a safe-call generation, the method further comprises:

training the speech generation model, comprising:

5. A method according to any one of claims 1 to 3, wherein prior to said screening selected speech template text from a pre-set insurance template database based on said risk category interest information, said method further comprises:

constructing the insurance template database, including:

6. The method according to claim 5, wherein the sentence screening of the candidate information sentences according to the preset sentence screening conditions to obtain the selected information sentences comprises:

7. The method of claim 5, wherein prior to populating the selected keywords with a preset information extraction template to obtain the candidate speech template text, the method further comprises:

filtering the selected keywords, including:

8. An insurance session generating device, the device comprising:

9. An electronic device comprising a memory storing a computer program and a processor implementing the method of generating an insurance session of any of claims 1 to 7 when the computer program is executed by the processor.

10. A computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method of generating an insurance policy according to any of claims 1 to 7.