CN119149699A

CN119149699A - Communication method and device thereof

Info

Publication number: CN119149699A
Application number: CN202411288957.0A
Authority: CN
Inventors: 陈峥
Original assignee: Vivo Mobile Communication Co Ltd
Current assignee: Vivo Mobile Communication Co Ltd
Priority date: 2024-09-14
Filing date: 2024-09-14
Publication date: 2024-12-17

Abstract

The application discloses a call method and a call device, and belongs to the technical field of interaction. The call method comprises the steps of determining a first character type of a call object, obtaining first character information of the call object, screening a first language model corresponding to the first character type from a first model set according to the first character type, determining a first prompt word based on the first character information, inputting the first prompt word into the first language model to obtain a first response result output by the first language model, and carrying out call with the call object based on the first response result.

Description

Communication method and device thereof

Technical Field

The application belongs to the technical field of interaction, and particularly relates to a conversation method and a conversation device.

Background

In the related technical scheme, a language model is deployed in the electronic equipment so as to replace a machine owner to communicate with a communication object by using the language model.

However, the language model deployed in the electronic device has a relatively fixed language style, and when the user is talking with the talking object instead of the owner, the attitude presented to the talking object is relatively hard, and it is difficult to dynamically represent the dynamic multifaceted character.

Disclosure of Invention

The embodiment of the application aims to provide a conversation method and a conversation device, which can solve the problem that the attitude presented to a conversation object is hard and the dynamic multifaceted character is difficult to dynamically represent when a language model deployed in electronic equipment replaces a machine owner to carry out conversation with the conversation object.

In a first aspect, an embodiment of the application provides a call method, which comprises the steps of determining a first character type of a call object, obtaining first character information of the call object, screening a first language model corresponding to the first character type from a first model set according to the first character type, determining a first prompt word based on the first character information, inputting the first prompt word into the first language model to obtain a first response result output by the first language model, and communicating with the call object based on the first response result.

In a second aspect, the embodiment of the application provides a call device, which comprises an acquisition module, a screening module, a determination module, a processing module and a response module, wherein the acquisition module is used for determining a first character type of a call object and acquiring first character information of the call object, the screening module is used for screening a first language model corresponding to the first character type from a first model set according to the first character type, the determination module is used for determining a first prompt word based on the first character information, the processing module is used for inputting the first prompt word into the first language model to obtain a first response result output by the first language model, and the response module is used for making a call with the call object based on the first response result.

In a third aspect, embodiments of the present application provide an electronic device comprising a processor and a memory storing a program or instructions executable on the processor, which when executed by the processor, implement the steps of the method as in the first aspect.

In a fourth aspect, embodiments of the present application provide a readable storage medium having stored thereon a program or instructions which when executed by a processor perform the steps of the method as in the first aspect.

In a fifth aspect, embodiments of the present application provide a chip comprising a processor and a communication interface, the communication interface being coupled to the processor, the processor being configured to execute programs or instructions to implement a method as in the first aspect.

In a sixth aspect, embodiments of the present application provide a computer program product stored in a storage medium, the program product being executable by at least one processor to implement a method as in the first aspect.

In the embodiment of the application, a call method is provided, in which a first character type of a call object and first character information of the call object can be determined, and a first language model corresponding to the first character type is screened out from a first model set by using the first character type, and in the process, the language model corresponding to the character type can be pertinently matched for the call object. Meanwhile, in the process of generating the first prompting word, the first character information of the call object is referred, so that the generated first prompting word is more adaptive to the call object, and the obtained first response result can reflect the dynamic multifaceted character, and further the call experience is improved.

Drawings

Fig. 1 shows one of the flow diagrams of the call method in the embodiment of the application;

FIG. 2 is a second flow chart of a call method according to an embodiment of the application;

FIG. 3 is a third flow chart illustrating a call method according to an embodiment of the application;

FIG. 4 is a flow chart of a call method according to an embodiment of the application;

FIG. 5 is a fifth flow chart of a call method according to an embodiment of the application;

FIG. 6 is a flowchart illustrating a call method according to an embodiment of the present application;

FIG. 7 is a schematic block diagram of a telephony device in an embodiment of the present application;

FIG. 8 shows a schematic block diagram of an electronic device in an embodiment of the application;

fig. 9 shows a schematic diagram of a hardware structure of an electronic device in an embodiment of the present application.

Detailed Description

Embodiments of the present application will be described more fully hereinafter with reference to the accompanying drawings, in which it is shown, however, in which some, but not all embodiments of the application are shown. All other embodiments, which are obtained by a person skilled in the art based on the embodiments of the present application, fall within the scope of protection of the present application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged, as appropriate, such that embodiments of the present application may be implemented in sequences other than those illustrated or described herein, and that the objects identified by "first," "second," etc. are generally of a type, and are not limited to the number of objects, such as the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/", generally means that the associated object is an "or" relationship.

The terms mentioned in the present application are explained below:

The large model refers to a large language model, is a deep learning model with super-large scale parameters and has semantic understanding and content generating capability, and the large language model is a deep learning model trained by using a large amount of text data, and can process various natural language tasks such as text classification, question-answering, dialogue and the like. The important form of the current intelligent agent is to take thinking and summarizing capability of a large language model as a core, connect external memory and tool components and realize the enhancement of intelligent capability.

Rewarding model the rewarding model is a model used for evaluating the behavior value of the agent in reinforcement learning and guides the learning of the agent by calculating the rewarding value of each interaction.

RLHF: RLHF refers to Reinforcement Learning fromHuman Feedback, human feedback reinforcement learning, which provides a natural and humanized interactive learning process for a machine by incorporating human feedback into the training process.

MBTI refers to a Myers-Buuges type index (Myers-Briggs Type Indicator, MBTI), and classifies personality into 16 types according to four dimensions of attention direction, cognition mode, judgment mode and life style.

Prompt in the artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) large model, the role of Prompt is primarily to Prompt the AI model for the context of the input information and the parameter information of the input model.

The multi-role large model set of the machine owner, namely the first model set in the application, namely the large language model set representing the thought characteristics of the machine owner, wherein each large language model in the set carries out optimization fine tuning on the dialogue target roles.

The following describes in detail the call method and the device provided by the embodiment of the application through specific embodiments and application scenarios thereof with reference to the accompanying drawings.

In one embodiment, as shown in fig. 1, a call method is proposed, which includes:

step 102, determining a first role type of the call object, and acquiring first role information of the call object.

In some embodiments of the present application, a conversation object may be understood as an object that interacts with an agent.

The call object may be an interactive call object, such as a friend of a machine owner, a chat object, such as a net friend of the machine owner, or a lecture object, such as a student of the machine owner.

In some embodiments of the present application, the first role type corresponding to the call object may be understood as a role type played by the call object in the call.

Illustratively, the first character type may be a friend, student, teacher, relative, or the like of the owner.

In some embodiments of the application, the role type range of the first role type is { parent, spouse, child, sibling, ancestor, leadership, subordinate, sibling, friend, other }.

In some embodiments of the present application, the first persona information of the call object includes evaluation information of the call object and call content generated by interaction with the call object.

The evaluation information of the call object can be understood as the evaluation condition of the call object by the machine owner, which can be obtained by the machine owner or obtained by counting historical interaction data.

Specifically, the evaluation information includes a trust degree and a sensitivity, wherein the trust degree refers to the trust degree of the owner on the call object, and the sensitivity refers to the initiative degree of the owner on the call object.

The trust level can be calculated by the following method:

Specifically, the confidence level is obtained by searching the historical communication record of the whole time period and the communication object on the electronic equipment, extracting the characteristics { first communication time, communication frequency, high-density information quantity and communication satisfaction }, and calculating the confidence level score I _trust by using a pre-trained confidence level calculation model.

The method for calculating the goodness comprises the steps of searching a historical communication record of a conversation object in the last month of equipment, extracting characteristics { communication frequency, conversation emotion tendency, active conversation duty ratio }, and calculating the goodness I _adore by using a pretrained goodness calculation model.

In some embodiments of the present application, the call content generated by the interaction with the call object may be understood as a substantial dialogue generated by the interaction with the call object.

Illustratively, the call object says feed, hello, and the owner replies to hello, the old classmate.

In some embodiments of the present application, the first character type may be obtained by identifying remark information of the call object with the history call record.

Specifically, if the recognition result is not "other", recording the character type R _cur of the corresponding interactive object, otherwise, using the "other" character large model in the first language model to inquire the identity information of the call object, and further obtaining the first character type.

Specifically, in the interaction process, the tone color of the interaction adopts the tone color adopted by the historical voice interaction of the owner, if the identity information answer of the call object accords with the preset character type value range, the corresponding character type R _cur is recorded, and otherwise, the character type is marked as 'other'.

Step 104, according to the first character type, screening a first language model corresponding to the first character type from the first model set.

In some embodiments of the present application, the first model set includes language models corresponding to at least one character type, and under the condition that the first character type is determined, the language models with the same character type as the first character type are searched in the first model set and used as the first language model.

Illustratively, the first character type is "teacher", and the language model with the character type of "teacher" is searched in the first model set as the first language model.

Step 106, determining a first prompt word based on the first character information.

Step 108, inputting the first prompt word into the first language model to obtain a first response result output by the first language model.

Step 110, a call is performed with the call object based on the first response result.

In the embodiment of the application, a call method is provided, in which a first character type corresponding to a call object and first character information of the call object can be determined, and a first language model corresponding to the call object is screened out from a first model set by using the first character type, and in the process, the language model corresponding to the character type can be pertinently matched for the call object. Meanwhile, in the process of generating the first prompting word, the first character information of the call object is referred, so that the generated first prompting word is more adaptive to the call object, and the obtained first response result can reflect the dynamic multifaceted character, and further the call experience is improved.

In some embodiments of the application, the call method further comprises the steps of identifying character trends of the machine owner based on historical use records of the electronic equipment to obtain character trend identification results, adjusting the first model based on the character trend identification results to obtain a second model, adjusting the second model based on the interaction data set to obtain a first model set, wherein the interaction data set comprises historical interaction data of different character types, and the first model set comprises language models corresponding to the different character types.

In some embodiments of the present application, the history of use record of the electronic device may be understood as a history of operation record of the electronic device, and the owner may interact with different call objects by operating the electronic device, or may fill in information by using the electronic device, and browse different pages.

In some embodiments of the present application, the first model may be understood as an untrained large model.

Illustratively, the first model is an initial large model.

In the above embodiment, the owner character tendency P _user belongs to the character tendency in the predefined character tendency set S _P, the character classification model is used to score the behaviors in the history usage record R _pi associated with each character, and the closest character tendency is selected as the owner character tendency P _user in the character tendency set S _P according to the average of the scores.

The history usage record of the character association comprises dialogue tuples { opposite side sentences, owner replies } set, and can also comprise a frequently-listened music style sequence { first music style, second music style, third music style }, or comprises note type statistics { lyric notes number, planning notes number }.

Specifically, MBTI sixteen personality can be used as the personality tendency set S _P, the personality classification model with confidence is used in the four personality dimensions of attention, cognition, judgment and life style, the history usage record R _pi of personality association is classified, and MBTI personality with the highest probability is obtained by combination as the owner personality tendency P _user.

In some embodiments of the present application, the character classification model may use a neural network, a support vector machine, or other common classification model as its structure.

In some embodiments of the present application, the second model LLM _PB is obtained according to the character trend recognition result P _user, that is, the second model LLM _PB is obtained according to the owner character trend P _user.

Wherein the large language model (Large Language Model, LLM), LLM _PB represents the second model.

Illustratively, the second model is a character base large model.

Specifically, for each character trend element in the predefined character trend set S _P, collecting the corpus conforming to the character features in the corpus set as training data, and performing full-scale fine tuning based on the first model LLM _B to obtain a second model set { LLM _PB-p }.

Wherein, P is the character trend element in the character trend set S _P, and P is the character trend recognition result P _user, so as to obtain the second model LLM _PB.

In the embodiment, a corresponding first model set can be created for the owner according to the history use record of the electronic equipment, and in the actual use process, the first response result is ensured to fully reflect the dynamic multifaceted performance of the owner, so that the interaction efficiency is improved.

In some embodiments of the application, the second model is adjusted based on the interaction data set to obtain the first model set, and the method specifically comprises the steps of determining a reward model set based on the interaction data set, wherein the reward model set comprises reward models corresponding to historical interaction data of different role types, and adjusting the second model based on the reward model set to obtain the first model set.

In this embodiment, the interaction data set, i.e., the multi-role dialogue data set { R _ci }, trains, for each dialogue object role i, a reward model Reward _ci for large model generation result evaluation, so as to evaluate the trained second model by using the reward model corresponding to the historical interaction data of different role types, and finally obtain a first model set capable of fully reflecting different role types, in this process, the language model in the first model set can be adapted to the corresponding role type, thereby obtaining a response result conforming to the dynamic multi-aspect of the character.

Specifically, the manner in which the reward model evaluates the second model generation result is to calculate the language style similarity of the second model generation result to the corresponding character dialogue dataset, and the similarity result is used as the evaluation score.

In some embodiments of the present application, the second model LLM _PB is tuned using the multi-role bonus model set { Reward _ci } by the method of RLHF.

Specifically, each character rewarding model carries out RLHF on the second model LLM _PB to obtain a main character large model LLM _Ci, i represents the character type, and RLHF is completed to obtain a main character multi-character large model set { LLM _Ci }.

In some embodiments of the application, a history usage record of an electronic device includes at least one piece of history interaction data, each piece of history interaction data has corresponding remark information, the conversation method further includes performing role classification on each piece of remark information by using a first classification model to obtain a target role type corresponding to each piece of history interaction data and a confidence coefficient corresponding to the target role type, taking the target role type as the role type of the corresponding history interaction data based on the confidence coefficient corresponding to the target role type being greater than or equal to a first threshold value, performing content recognition on the history interaction data based on the confidence coefficient corresponding to the target role type being less than the first threshold value to obtain the role type corresponding to the history interaction data, and constructing an interaction data set based on the history interaction data and the role type corresponding to the history interaction data.

In this embodiment, the historical interaction data may be one or more of historical phone call data, historical text message data, and historical chat records of chat software.

Specifically, by setting a time window T, the chat record is cut to obtain dialogue data segments, i.e., historical interaction data.

In some embodiments of the present application, the remark information is a remark for the call object.

Illustratively, the remark information is "mother in love" or "sister in school".

Each piece of remark information is subjected to role classification by adopting a first classification model so as to attempt to acquire the obtained role type, and a first threshold value is set so as to evaluate the confidence degree of the target role type acquired by classification by using the first classification model.

Specifically, if the confidence coefficient corresponding to the target role type is greater than or equal to the first threshold, the reliability of the target role type is considered to be relatively high, the target role type can be used as the role type of the corresponding historical interaction data, otherwise, the historical interaction data is subjected to content recognition, and the role type corresponding to the historical interaction data is determined through the content recognition.

In the process, the efficiency of determining the role type can be improved, and meanwhile, the accuracy of determining the role type is improved.

For example, if the first threshold value is 0.8, the confidence coefficient corresponding to the target role type is greater than or equal to 0.8, and if the confidence coefficient corresponding to the target role type is less than 0.8, the content recognition is performed on the historical interaction data to obtain the role type corresponding to the historical interaction data.

In general, different remark information is set for different call objects, so that the role types corresponding to the historical interaction data are obtained by using the remark information, the accuracy of the role types corresponding to the historical interaction data can be ensured, and meanwhile, the efficiency of determining the role types corresponding to the historical interaction data can be improved.

And the content of the historical interaction data is identified, so that the role type corresponding to the historical interaction data can be accurately obtained under the condition that the classification by using the first classification model is unreasonable, and the use requirements under different scenes can be met.

In some embodiments of the application, the content recognition is performed on the historical interaction data to obtain the role types corresponding to the historical interaction data, and the method specifically comprises the steps of segmenting chat contents according to a time window, extracting topics from each segment of chat contents by using a large language model, and obtaining a topic sequence.

And carrying out frequency statistics on topic elements in the topic sequence to obtain topic distribution D _chatt, and comparing preset chat topic distribution of each role type to obtain the role type with highest similarity. If the similarity between the topic distribution of the chat content and the topic distribution of the role type is greater than a threshold Th _c, determining that the current role type is the role type, otherwise, taking the role type value as 'other'.

Where the historical interaction data is data in a multi-persona dialog dataset { R _ci }, illustratively, the multi-persona dialog dataset { R _ci } includes a binary group { dialog object role type, dialog data segment }.

In some embodiments of the application, the first classification model is a pre-trained remark name classification model.

In some embodiments of the application, when the first character information of the call object comprises evaluation information of the call object and call content generated by interaction with the call object, the evaluation information comprises trust degree and goodness, the first prompt word is determined based on the first character information, the memory prompt word is determined based on the first character type, the trust degree and the personalized memory data block, the scene prompt word is determined based on the first character type and the goodness, the historical conversation prompt word is determined based on the call content, the main information prompt word determined based on the main description information is acquired, the output requirement prompt word is acquired, and the first prompt word is determined based on the memory prompt word, the scene prompt word, the historical conversation prompt word, the main information prompt word and the output requirement prompt word.

In this embodiment, the memory hint may be understood as a hint determined based on historical interaction data of the call object, the scene hint may be understood as a hint determined based on historical interaction impressions of the call object, the historical session hint may be understood as a hint determined by referencing the interaction content and the personalized memory data block, the owner information hint may be understood as a hint related to the owner, and the output requirement hint may be understood as a hint set to the language model output format.

Wherein the output requirement Prompt _req is a Prompt generated from a template of the language model output format.

Illustratively, [ output Requirements ] please answer in the following format:

{

whether the current session can be terminated: ".

Whether the current counterpart initiates a decision problem, what the content of the problem is% "

Your answer: "".

}

The personalized memory data block is understood to be a data block determined by historical interaction data of a call object. The use of the personalized memory data block can enable the determined historical conversation prompt word to relate to the historical interaction data of the call object, so that the accuracy of the first prompt word is improved.

The owner description information may be understood as information for describing the owner.

In this embodiment, the accuracy of the first response result output by the first language model may be improved by generating the first prompt word based on different prompt words.

Illustratively, if the owner description information is represented by a person-set information key value pair Prof, the owner description information is serialized by using a preset format character, and a owner information Prompt word, that is, a person-set Prompt _Prof is generated.

Illustratively, the generated person sets Prompt _Prof to "you are a civil engineer, male, 30 years old, like to play a table tennis".

In some embodiments of the application, based on the first character type, the trust degree and the personalized memory data block, the memory Prompt word is determined, and the method specifically comprises the steps of searching the personalized memory data block with high association degree with the dialogue content Q in the personalized memory data M, screening objects meeting control conditions { the dialogue object character type R _cur and the trust degree score I _trust } by using authority control labels associated with the data block, and obtaining the memory Prompt word of the associated dialogue, wherein the memory Prompt word is the personalized memory Prompt word Prompt _M.

Illustratively, for dialogue content Q "what was done yesterday", the memory hint word is "[ memory ] you play a table tennis at gym yesterday.

In some embodiments of the present application, the historical conversation hint word Prompt _hig is determined based on conversation content.

The call content, namely the history record generated by the interaction.

Illustratively, "[ historical dialog ] is your dialog record:

[ opposite side ] feed, hello.

[ You ]: hello, old classmate.

The first prompting word can be a prompting word obtained by sequentially connecting a memory prompting word, a scene prompting word, a history session prompting word, a machine owner information prompting word and an output requirement prompting word.

In some embodiments of the present application, the first prompting word is a prompting word obtained by sequentially connecting a machine owner information prompting word, a memory prompting word, a scene prompting word, a history session prompting word and an output requirement prompting word.

In some embodiments of the application, the call method further comprises generating personalized memory data based on the historical usage record of the electronic device, wherein the personalized memory data comprises owner description information and personalized memory data blocks.

In this embodiment, the personalized memory data may be pre-constructed so that the first hint word is determined based on the personalized memory data.

In some embodiments of the application, personalized memory data is generated based on the historical usage record of the electronic device, and specifically comprises the steps of determining owner description information based on owner personal information and owner preference information filled in the historical usage record of the electronic device, and determining a personalized memory data block based on an interaction data set determined by the historical usage record of the electronic device.

The personal information of the owner, namely owner setting information, may be form filling history data of the owner in a highly reliable software environment such as a bank, a institution and the like, search and extract objective condition person setting information, and record person setting information meeting a preset objective condition person setting project table W _opro.

Optionally, the objective condition person sets the project table W _opro to include name, gender, age, through, and residence information.

In some embodiments of the present application, the owner preference information may obtain the activity type of the owner high frequency software usage scenario based on the owner history software usage frequency statistics, and generate preference class setting information.

The objective condition people setting information and the fan setting information form owner description information together, namely a person setting information key value pair Prof.

In some embodiments of the present application, the interaction data set, that is, the personalized memory data block set { M _ci }, integrally records the historical interaction data with the information entropy greater than the preset threshold value as the personalized memory data block M _ci by calculating the information entropy of the historical interaction data, where i represents the character type.

To the personalized memory data block M _ci, a personalized data block permission label { MPerm _ci } is added. And each personalized memory data block is provided with a right control tag corresponding to the full role type, and the right control tag consists of a binary group { role type, lowest accessible trust degree }, and the content of the personalized memory data block can be accessed only when the trust degree reaches the lowest accessible trust degree during the conversation with personnel corresponding to the role type.

The trust value in the personalized data block authority label is generated by the big model according to the personalized memory data block content and role type decision.

After adding the personalized data block permission label { MPerm _ci }, vectorizing the personalized memory data and warehousing.

Specifically, the personalized data block with the authority label is stored in a vector database, and the generation of the queriable personalized memory data M is completed.

In some embodiments of the application, a call is performed with a call object based on a first response result, specifically comprising the steps of analyzing the first response result to obtain an analysis result, determining the problem content of a decision problem based on the first response result when the call is not ended and the decision problem exists based on the analysis result, determining a second prompt word based on the problem content of the decision problem, a first character type and evaluation information of the call object, inputting the second prompt word into a first language model to obtain a second response result output by the first language model, and performing the call with the call object based on the second response result.

In this embodiment, a new prompt word can be generated under the condition of having a decision problem, and the new prompt word, that is, the second prompt word, is input into the first language model to obtain a response result of the first language model, and in this process, the decision problem can be processed, so that the controllability of the call response result is increased.

Illustratively, for a "colleague" character type that mentions a need to borrow money in a conversation, the corresponding decision problem content "borrow money" is generated, the output of the character big model may be:

{

whether the decision problem agrees with "unable to determine"

Deciding whether the problem generates a task to be handled, namely, judging whether to borrow money for colleagues;

the answer content of your dialogue is "I and at home business, give you answer at night"

}

Wherein, the second response result in the application is obtained by 'i and home business take business and give you a reply at a later time'.

The splicing strategy of the second prompting words is as follows:

You are talking with { { { talking object role type R _cur }, the trust of the other party is { { trust I _trust }, the goodness is { { goodness I _adore }, and for the decision problem { { { decision problem content D }, you agree or not;

please answer in the following format, wherein the answer to decide if the question agrees is a choice of "yes, no, indeterminate";

{

decision whether the question agrees'

Decision whether the question generates a task to be done "

Your dialogue answer content: ".

}

In some embodiments of the present application, when it is determined that the call is not ended and there is no decision problem based on the analysis result, a first response result is output.

In this embodiment, when it is determined based on the analysis result that the call is not ended and there is no decision problem, the first response result is directly output to meet the call response requirement in the case of no decision problem.

In some embodiments of the present application, taking a call as an example, as shown in fig. 2, the call method includes:

step 202, a first set of models is generated from a device usage record.

Wherein the multi-role large model set { LLM _Ci } is the first model set in the present application.

Step 204, generating personalized memory data M according to the device usage record.

And 206, performing call response based on the owner multi-role large model set { LLM _Ci } and the personalized memory data M.

As shown in fig. 3, in step 202, a first model set is generated according to a device usage record, specifically including:

step 302, identifying the character trend of the machine according to the device use record.

And step 304, acquiring a second model according to the character trend identification result.

Step 306, classifying the historical dialogue record to obtain an interactive data set.

Step 308, generating a multi-role rewards model set according to the interactive data set.

And 310, tuning the second model by using the multi-role rewarding model set in a RLHF mode to obtain a first model set.

In step 306, as shown in fig. 4, the historical dialogue records are classified to obtain an interaction data set, which specifically includes:

Step 402, attempting to acquire a character type according to remark information.

In step 404, the remark information is successfully obtained from the character type, and if yes, the process is ended, and if no, the process proceeds to step 406.

Step 406, obtaining the character type according to the chat content.

In step 204, as shown in fig. 5, the personalized memory data M is generated according to the device usage record, which specifically includes:

Step 502, owner setup information extraction.

Step 504, personalized memory database extraction.

Step 506, personalized data block rights label addition.

Step 508, vectorizing and warehousing the personalized memory data.

In step 206, as shown in fig. 6, a call response is performed based on the owner multi-role large model set { LLM _Ci } and the personalized memory data M, which specifically includes:

Step 602, identification of the role type of the call object.

Step 604, call object relationship identification.

Step 606, call content extraction.

Step 608, character big model prompt word generation.

Step 610, role large model dialog query.

Step 612, if the dialogue is ended, step 620 is executed if the judgment result is yes, and step 614 is executed if the judgment result is no.

Step 614, if there is a decision problem, step 616 is executed if the decision is yes, and step 618 is executed if the decision is no.

Step 616, decision problem handling.

And 618, outputting a call reply language.

Step 620, end the call.

Step 622, call summary.

In call summary, main content, decision records and tasks to be handled are summarized by using a role large model according to the round of call records.

Optionally, the prompt word input to the character big model is "please help summarize main content, decision record and task to be done according to the call record.

According to the communication method provided by the embodiment of the application, the execution main body can be a communication device. In the embodiment of the present application, a call device executing a call method is taken as an example, and the call device provided in the embodiment of the present application is described.

In some embodiments of the present application, as shown in fig. 7, a call device 700 is provided, which includes an obtaining module 702 configured to determine a first character type of a call object and obtain first character information of the call object, a screening module 704 configured to screen a first language model corresponding to the first character type from a first model set according to the first character type, a determining module 706 configured to determine a first prompt word based on the first character information, a processing module 708 configured to input the first prompt word to the first language model to obtain a first response result output by the first language model, and a response module 710 configured to make a call with the call object based on the first response result.

In the embodiment of the application, the first character type corresponding to the call object and the first character information of the call object can be obtained, and the first language model corresponding to the first character type can be screened out from the first model set by utilizing the first character type. Meanwhile, in the process of generating the first prompting word, the first character information is referred to, so that the generated first prompting word is more adaptive to a call object, and the obtained first response result can reflect the dynamic multifaceted character, and further the call experience is improved.

In some embodiments of the present application, the processing module 708 is further configured to identify a character trend of the machine owner based on a history of use of the electronic device to obtain a character trend identification result, adjust the first model based on the character trend identification result to obtain a second model, adjust the second model based on the interaction data set to obtain a first model set, where the interaction data set includes historical interaction data of different character types, and the first model set includes language models corresponding to the different character types.

In some embodiments of the present application, the processing module 708 is specifically configured to determine a reward model set based on the interaction data set, where the reward model set includes reward models corresponding to historical interaction data of different character types, and adjust the second model based on the reward model set to obtain the first model set.

In some embodiments of the present application, the history usage record of the electronic device includes at least one piece of history interaction data, each piece of history interaction data has corresponding remark information, and the processing module 708 is further configured to perform role classification on each piece of remark information by using a first classification model to obtain a target role type corresponding to each piece of history interaction data and a confidence coefficient corresponding to the target role type, take the target role type as a role type of the corresponding history interaction data based on the confidence coefficient corresponding to the target role type being greater than or equal to a first threshold, perform content recognition on the history interaction data based on the confidence coefficient corresponding to the target role type being less than the first threshold, and construct an interaction data set based on the history interaction data and the role type corresponding to the history interaction data.

In some embodiments of the present application, in the case where the first persona information of the call object includes evaluation information of the call object and call content generated by interaction with the call object, the evaluation information includes a confidence level and a sensitivity level, the determining module 706 is specifically configured to determine a memory hint word based on the first persona type, the confidence level, and the personalized memory data block, determine a scene hint word based on the first persona type and the sensitivity level, determine a historical conversation hint word based on the call content, obtain a principal information hint word determined based on the principal description information, and obtain an output requirement hint word, and determine a first hint word based on the memory hint word, the scene hint word, the historical conversation hint word, the principal information hint word, and the output requirement hint word.

In some embodiments of the present application, the processing module 708 is further configured to generate personalized memory data based on the historical usage record of the electronic device, wherein the personalized memory data includes owner descriptive information and a personalized memory data block.

In some embodiments of the present application, the processing module 708 is specifically configured to determine owner descriptive information based on owner personal information and owner preference information filled in a historical usage record of the electronic device, and determine a personalized memory data block based on the set of interaction data determined by the device usage record.

In some embodiments of the present application, the response module 710 is specifically configured to parse the first response result to obtain a parsed result, determine, based on the parsed result, a problem content of the decision problem based on the first response result when the call is not ended and there is a decision problem, determine a second prompt word based on the problem content of the decision problem, the first character type and the evaluation information of the call object, input the second prompt word into the first language model to obtain a second response result output by the first language model, and call with the call object based on the second response result.

In some embodiments of the present application, the answering module 710 is specifically further configured to output a first response result when it is determined that the call is not ended and there is no decision problem based on the parsing result.

The communication device in the embodiment of the application can be an electronic device or a component in the electronic device, such as an integrated circuit or a chip. The electronic device may be a terminal, or may be other devices than a terminal. The electronic device may be a Mobile phone, a tablet computer, a notebook computer, a palm computer, a vehicle-mounted electronic device, a Mobile internet appliance (Mobile INTERNET DEVICE, MID), an augmented reality (augmented reality, AR)/Virtual Reality (VR) device, a robot, a wearable device, an ultra-Mobile personal computer (UMPC), a netbook or a Personal Digital Assistant (PDA), etc., and may also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a Television (TV), a teller machine, a self-service machine, etc., which are not particularly limited in the embodiments of the present application.

The communication device in the embodiment of the application can be a device with an operating system. The operating system may be an Android operating system, an IOS operating system, or other possible operating systems, and the embodiment of the present application is not limited specifically.

The call device provided by the embodiment of the present application can implement each process implemented by the call method embodiment of fig. 1, and can achieve the same technical effects, and in order to avoid repetition, the description is omitted here.

As shown in fig. 8, the embodiment of the present application further provides an electronic device 800, which includes a processor 802 and a memory 804, where the memory 804 stores a program or instructions that can be executed on the processor 802, and the program or instructions implement the steps of the foregoing embodiments of the communication method when executed by the processor 802, and achieve the same technical effects, so that repetition is avoided and redundant description is omitted herein.

The electronic device in the embodiment of the application includes the mobile electronic device and the non-mobile electronic device.

Fig. 9 is a schematic hardware structure of an electronic device implementing an embodiment of the present application.

As shown in fig. 9, the electronic device 900 includes, but is not limited to, a radio frequency unit 901, a network module 902, an audio output unit 903, an input unit 904, a sensor 905, a display unit 906, a user input unit 907, an interface unit 908, a memory 909, and a processor 910.

Those skilled in the art will appreciate that the electronic device 900 may also include a power source (e.g., a battery) for powering the various components, which may be logically connected to the processor 910 by a power management system to perform functions such as managing charge, discharge, and power consumption by the power management system. The electronic device structure shown in fig. 9 does not constitute a limitation of the electronic device, and the electronic device may include more or less components than shown, or may combine certain components, or may be arranged in different components, which are not described in detail herein.

In some embodiments of the present application, the processor 910 is configured to determine a first character type of a call object and obtain first character information of the call object, screen a first language model corresponding to the first character type from a first model set according to the first character type, determine a first prompt word based on the first character information, input the first prompt word to the first language model to obtain a first response result output by the first language model, and make a call with the call object based on the first response result.

In some embodiments of the present application, the processor 910 is further configured to identify a character trend of the machine owner based on a history of use of the electronic device to obtain a character trend identification result, adjust the first model based on the character trend identification result to obtain a second model, adjust the second model based on the interaction data set to obtain a first model set, where the interaction data set includes historical interaction data of different character types, and the first model set includes language models corresponding to the different character types.

In some embodiments of the present application, the processor 910 is specifically configured to determine a reward model set based on the interaction data set, where the reward model set includes reward models corresponding to historical interaction data of different character types, and adjust the second model based on the reward model set to obtain the first model set.

In some embodiments of the present application, the history usage record of the electronic device includes at least one piece of history interaction data, each piece of history interaction data has corresponding remark information, the processor 910 is further configured to perform role classification on each piece of remark information by using a first classification model to obtain a target role type corresponding to each piece of history interaction data and a confidence coefficient corresponding to the target role type, use the target role type as a role type of the corresponding history interaction data based on the confidence coefficient corresponding to the target role type being greater than or equal to a first threshold, perform content recognition on the history interaction data based on the confidence coefficient corresponding to the target role type being less than the first threshold, and construct an interaction data set based on the history interaction data and the role type corresponding to the history interaction data.

In some embodiments of the present application, the processor 910 is specifically configured to parse the first response result to obtain a parsed result, determine, based on the parsed result, a problem content of the decision problem based on the first response result when the call is not ended and there is a decision problem, determine a second prompt word based on the problem content of the decision problem, the first character type and the evaluation information of the call object, input the second prompt word into the first language model to obtain a second response result output by the first language model, and call with the call object based on the second response result.

It should be appreciated that in embodiments of the present application, the input unit 904 may include a graphics processor (Graphics Processing Unit, GPU) 9041 and a microphone 9042, with the graphics processor 9041 processing image data of still pictures or video obtained by an image capture device (e.g., a camera) in a video capture mode or an image capture mode. The display unit 906 may include a display panel 9061, and the display panel 9061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 907 includes at least one of a touch panel 9071 and other input devices 9072. Touch panel 9071, also referred to as a touch screen. The touch panel 9071 may include two parts, a touch detection device and a touch controller. Other input devices 9072 may include, but are not limited to, a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and so forth, which are not described in detail herein.

The memory 909 may be used to store software programs as well as various data. The memory 909 may mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area may store an operating system, application programs or instructions (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like. Further, the memory 909 may include a volatile memory or a nonvolatile memory, or the memory 909 may include both volatile and nonvolatile memories. The nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable EPROM (EEPROM), or a flash Memory. The volatile memory may be random access memory (Random Access Memory, RAM), static random access memory (STATIC RAM, SRAM), dynamic random access memory (DYNAMIC RAM, DRAM), synchronous Dynamic Random Access Memory (SDRAM), double data rate Synchronous dynamic random access memory (Double DATA RATE SDRAM, DDRSDRAM), enhanced Synchronous dynamic random access memory (ENHANCED SDRAM, ESDRAM), synchronous link dynamic random access memory (SYNCH LINK DRAM, SLDRAM), and Direct random access memory (DRRAM). Memory 909 in embodiments of the application includes, but is not limited to, these and any other suitable types of memory.

Processor 910 may include one or more processing units, and optionally, processor 910 integrates an application processor that primarily processes operations involving an operating system, user interface, application program, etc., and a modem processor that primarily processes wireless communication signals, such as a baseband processor. It will be appreciated that the modem processor described above may not be integrated into the processor 910.

The embodiment of the application also provides a readable storage medium, and the readable storage medium stores a program or an instruction, which when executed by a processor, implements each process of the above-mentioned call method embodiment, and can achieve the same technical effects, so that repetition is avoided, and no further description is provided here.

The processor is a processor in the electronic device in the above embodiment. Readable storage media include computer readable storage media such as computer readable memory ROM, random access memory RAM, magnetic or optical disks, and the like.

The embodiment of the application further provides a chip, the chip comprises a processor and a communication interface, the communication interface is coupled with the processor, the processor is used for running programs or instructions, the processes of the communication method embodiment can be realized, the same technical effects can be achieved, and the repetition is avoided, and the description is omitted here.

It should be understood that the chips referred to in the embodiments of the present application may also be referred to as system-on-chip chips, chip systems, or system-on-chip chips, etc.

Embodiments of the present application provide a computer program product stored in a storage medium, where the program product is executed by at least one processor to implement the respective processes of the foregoing embodiments of the call method, and achieve the same technical effects, and for avoiding repetition, a detailed description is omitted herein.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Furthermore, it should be noted that the scope of the methods and apparatus in the embodiments of the present application is not limited to performing the functions in the order shown or discussed, but may also include performing the functions in a substantially simultaneous manner or in an opposite order depending on the functions involved, e.g., the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the embodiments of the present application may be embodied in essence or contributing to the prior art in the form of a computer software product stored on a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a cell phone, computer, server, or network device, etc.) to perform the methods of the various embodiments of the present application.

The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are to be protected by the present application.

Claims

1. A method of communicating, comprising:

Determining a first role type of a call object, and acquiring first role information of the call object;

Screening a first language model corresponding to the first character type from a first model set according to the first character type;

determining a first prompt word based on the first character information;

inputting the first prompt word into the first language model to obtain a first response result output by the first language model;

And communicating with the communicating object based on the first response result.

2. The method of claim 1, the communication method is characterized by further comprising the following steps:

Identifying character trends of the machine owner based on the history use record of the electronic equipment to obtain character trend identification results;

adjusting the first model based on the character trend recognition result to obtain a second model;

And adjusting the second model based on the interaction data set to obtain the first model set, wherein the interaction data set comprises historical interaction data of different role types, and the first model set comprises language models corresponding to the different role types.

3. The method for speaking according to claim 2, wherein said adjusting the second model based on the interaction data set to obtain the first model set specifically includes:

Determining a rewarding model set based on the interaction data set, wherein the rewarding model set comprises rewarding models corresponding to historical interaction data of different role types;

and adjusting the second model based on the rewarding model set to obtain the first model set.

4. The method of claim 2, wherein the historical usage record of the electronic device includes at least one piece of the historical interaction data, each piece of the historical interaction data having corresponding remark information, the method further comprising:

Performing role classification on each piece of remark information by adopting a first classification model to obtain a target role type corresponding to each piece of historical interaction data and a confidence coefficient corresponding to the target role type;

Based on the confidence coefficient corresponding to the target role type being greater than or equal to a first threshold, taking the target role type as the role type of the corresponding historical interaction data;

based on the confidence coefficient corresponding to the target role type is smaller than the first threshold, carrying out content identification on the historical interaction data to obtain the role type corresponding to the historical interaction data;

and constructing the interaction data set based on the historical interaction data and the role type corresponding to the historical interaction data.

5. The method according to any one of claims 1 to 4, wherein the communicating with the communicating object based on the first response result specifically includes:

analyzing the first response result to obtain an analysis result;

determining the problem content of the decision problem based on the first response result when the conversation is not ended and the decision problem exists based on the analysis result;

determining a second prompt word based on the problem content of the decision problem, the first character type and the evaluation information of the call object;

Inputting the second prompt word into the first language model to obtain a second response result output by the first language model;

And communicating with the communicating object based on the second response result.

6. A telephony device, comprising:

The system comprises an acquisition module, a judgment module and a display module, wherein the acquisition module is used for determining a first role type of a call object and acquiring first role information of the call object;

the screening module is used for screening a first language model corresponding to the first character type from a first model set according to the first character type;

The determining module is used for determining a first prompt word based on the first character information;

The processing module is used for inputting the first prompt word into the first language model to obtain a first response result output by the first language model;

and the response module is used for communicating with the communication object based on the first response result.

7. A call apparatus according to claim 6, wherein,

The processing module is further configured to:

8. The communication device according to claim 7, wherein the processing module is specifically configured to:

9. The telephony device of claim 7, wherein the historical usage record for the electronic device comprises at least one piece of the historical interaction data, each piece of the historical interaction data having corresponding remark information, the processing module further configured to:

10. The telephony device of any one of claims 6 to 9, wherein the answering module is configured to:

analyzing the first response result to obtain an analysis result;