CN117520503A - Financial customer service dialogue generation method, device, equipment and medium based on LLM model - Google Patents
Financial customer service dialogue generation method, device, equipment and medium based on LLM model Download PDFInfo
- Publication number
- CN117520503A CN117520503A CN202311494476.0A CN202311494476A CN117520503A CN 117520503 A CN117520503 A CN 117520503A CN 202311494476 A CN202311494476 A CN 202311494476A CN 117520503 A CN117520503 A CN 117520503A
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- vector
- target
- financial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 239000013598 vector Substances 0.000 claims abstract description 308
- 238000004590 computer program Methods 0.000 claims description 20
- 230000003993 interaction Effects 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 230000015572 biosynthetic process Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000003786 synthesis reaction Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
Abstract
The invention discloses a financial customer service dialogue generation method, a device, a computer device and a storage medium based on an LLM model, wherein the dialogue generation method comprises the following steps: acquiring a target problem, and converting the target problem into a target problem vector; according to the target problem vector, N similar problem vectors are obtained in a financial problem vector library in a searching mode, wherein N is more than 2; according to the identification of the financial service scene, sorting N similar problem vectors, and selecting M similar problem vectors matched with the identification of the financial service scene, wherein M is less than N; according to the M similar question vectors, retrieving M question-answer vector groups from a question-answer database; generating a scene question-answer prompt template according to the target question vector, the financial service scene identifier and M question-answer vector groups; and inputting the scene question-answer prompt template into the LLM model, and outputting target answer text information. The method improves the accuracy of the financial customer service dialogue.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to a financial customer service dialogue generation method, device, equipment and medium based on an LLM model.
Background
In the existing financial insurance intelligent customer service system, call records, online question and answer records and other information of the existing customer service system are cleaned and marked with question and answer pairs by manpower, the question and answer pairs are arranged by manpower, then similar questions and the question and answer pairs are respectively stored in a similar question vector library and a relational database for standby, when a user accesses the financial insurance intelligent customer service system, the similar questions are obtained by searching the similar question vector library, and answers are returned to the user in a similar question query relational database mode. In the method, a large amount of manpower is generally required to maintain and update question-answer pairs in the earlier storage period of the similar question vector library and the relational database, and when answer information is returned to a user, only answers of the question-answer pairs stored in the database can be returned, and the answers of the question-answer pairs are possibly not answers actually required by the user and cannot reasonably solve the requirements of the user.
Disclosure of Invention
Based on the method, the device, the equipment and the medium for generating the financial customer service dialogue based on the LLM model are provided, so that the problems that the database maintenance consumes a great deal of labor cost and the customer appeal cannot be solved comprehensively and accurately in the prior art are solved.
In a first aspect, an embodiment of the present invention provides a method for generating a financial customer service session based on LLM model, the method including the steps of:
acquiring a target problem, and converting the target problem into a target problem vector;
according to the target problem vector, N similar problem vectors are obtained through searching in a preset financial problem vector library, wherein N is more than 2;
according to a preset financial service scene identifier, sorting N similar problem vectors, and selecting to obtain M similar problem vectors matched with the financial service scene identifier, wherein M is less than N;
according to the M similar question vectors, retrieving M question-answer vector groups from a preset question-answer database, wherein each question-answer vector group comprises a standard question vector and a standard answer;
generating a scene question-answer prompt template according to the target question vector, the financial service scene identifier and M question-answer vector groups;
and inputting the scene question-answering prompt template into the LLM model which is trained in the question-answering of the financial business, and outputting target answer text information.
In a second aspect, an embodiment of the present invention provides a device for generating a financial customer service session based on LLM model, the device comprising:
The target problem conversion module is used for acquiring a target problem and converting the target problem into a target problem vector;
the similar problem retrieval module is used for retrieving N similar problem vectors in a preset financial problem vector library according to the target problem vector, wherein N is more than 2;
the similar problem ordering module is used for ordering the N similar problem vectors according to a preset financial service scene identifier, and selecting M similar problem vectors matched with the financial service scene identifier, wherein M is smaller than N;
the question-answer pair retrieval module is used for retrieving M question-answer pair vector groups from a preset question-answer database according to the M similar question vectors, wherein each question-answer pair vector group comprises a standard question vector and a standard answer vector;
the question-answer prompting template generation module is used for generating a scene question-answer prompting template according to the target question vector, the financial service scene identifier and M question-answer vector groups;
and the model output module is used for inputting the scene question-answer prompt template into the LLM model which is trained by the financial business question-answer, and outputting target answer text information.
In a third aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the steps of the LLM model-based financial customer service session generation method described above when the processor executes the computer program.
In a fourth aspect, an embodiment of the present invention provides a computer readable storage medium storing a computer program, where the computer program when executed by a processor implements the steps of the LLM model-based financial customer service session generation method described above.
In a fifth aspect, an embodiment of the present invention provides a computer program product, which when executed on a terminal device, causes the terminal device to execute the steps of the method for generating a financial customer service session based on LLM model.
According to the method, the device, the equipment and the medium for generating the financial customer service dialogue based on the LLM model, the target questions are converted into the target question vectors, N similar question vectors are obtained by searching in a financial question vector library, M similar question vectors matched with the preset financial service scene identifications are selected from the N similar question vectors, M question-answer vector groups are obtained by searching in a question-answer database according to the M similar question vectors, a scene question-answer prompt template comprising the target question vectors, the financial service scene identifications and the M question-answer vector groups is generated, the scene question-answer prompt template is input into the LLM model which is subjected to the financial service question-answer training, and target answer text information is output.
Compared with the existing financial insurance intelligent customer service system, the intelligent customer service system takes the scene question-answer prompt template containing insurance background information as the input of the LLM model, the LLM model improves the accuracy of target answer information by understanding and learning the known question-answer in the scene question-answer prompt template to prompt information, and on the basis, the LLM model carries out reasoning and expansion on the learned question-answer to prompt information by combining language knowledge obtained during model pre-training, generates a plurality of candidate answers, selects the most suitable answer and returns the most suitable answer to a user, improves the comprehensiveness of the target answer information, and the target answer text information and the target question generated by the LLM model can form a new question-answer pair, so that the expansion of the question-answer pair information in a database is realized, and the cost of manpower maintenance question-answer pair database is effectively reduced.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of an application environment of a LLM model-based financial customer service session generation method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for generating a financial customer service session based on LLM model according to an embodiment of the present invention;
FIG. 3 is a flow chart of a method for generating a financial customer service session based on LLM model for retrieving similar problem vectors according to an embodiment of the present invention;
FIG. 4 is a flow chart of matching similar problem vectors in a LLM model-based financial customer service session generation method according to an embodiment of the present invention;
FIG. 5 is a flow chart of a model output target answer text message in a LLM model-based financial customer service session generation method according to an embodiment of the present invention;
FIG. 6 is a flow chart of retrieving and sending question-answer pair vector sets in a LLM model-based financial customer service dialogue generation method according to an embodiment of the invention;
FIG. 7 is a schematic block diagram of a LLM model-based financial customer service session generation device according to an embodiment of the present invention;
FIG. 8 is a schematic diagram of a computer device in accordance with an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The financial customer service dialogue generation method based on the LLM model can be applied to an application environment as shown in figure 1, wherein a client (computer equipment) communicates with a server through a network. The client acquires the financial customer service dialogue request and sends the financial customer service dialogue request to the server. After the server side obtains the financial customer service dialogue request, the server side processes the financial customer service dialogue request correspondingly and responds to the financial customer service dialogue request. The computer device may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, among others. The server may be implemented by a stand-alone server or a server cluster formed by a plurality of servers.
In an embodiment, as shown in fig. 2, a method for generating a financial customer service session based on LLM model is provided, and the method is applied to the server in fig. 1 for illustration, and includes the following steps:
s10: and acquiring a target problem, and converting the target problem into a target problem vector.
Specifically, after the target question is acquired, the target question may be converted into a target question vector by a word embedding technique. The target question text is divided into text blocks by using a pre-trained word embedding model, such as text2vec, ernie or m3e, and each text block is mapped to a specific position in a high-dimensional vector space to obtain a target question vector, namely the conversion from the target question text to the target question vector is completed. Where each text block may represent context semantic information for each text block at a particular location in the high-dimensional vector space, e.g., text blocks with similar meanings may be more closely related in the vector space.
S20: and according to the target problem vector, N similar problem vectors are retrieved from a preset financial problem vector library, wherein N is more than 2.
The financial problem vector library refers to a database for storing and managing all similar problem vectors in the financial field, wherein each similar problem vector stored in the financial problem vector library has a corresponding data structure, the data structure may represent a corresponding relationship among each similar problem vector, a similar problem sentence and each question-answer pair vector group, and a specific form of the data structure may be set according to actual situations, for example, the data structure may be: similarity problem vector: sentence of similar problem: according to the data structure, the corresponding similar problem sentences can be found through the similar problem vectors, and through the coding of the similar problem vectors, the corresponding question-answer vector groups can be retrieved from a question-answer database.
Specifically, N similar problem vectors may be retrieved from a preset financial problem vector database by constructing a vector index. Firstly, an index structure of similar problem vectors is established, vectors with the same financial service type as the target problem in a financial problem vector library are loaded into the similar problem vector index structure in a matrix form, secondly, the target problem vectors to be searched are added into the established similar problem vector index structure, a plurality of target problem vectors can be added at one time or one by one, again, the similarity between each similar problem vector and the target problem vector is calculated, and a similarity measurement standard between vectors is set to obtain the similar problem vectors meeting the similarity measurement standard, the standard can be adjusted according to actual conditions, and finally N similar problem vectors reaching the similarity measurement standard between the target problem vectors are output.
S30: and sequencing the N similar problem vectors according to a preset financial service scene identifier, and selecting to obtain M similar problem vectors matched with the financial service scene identifier, wherein M is smaller than N.
In the insurance development business type, the preset financial business scene may include a pre-insurance consultation scene and a post-insurance sales service scene, in the underwriting business type, the preset financial business scene may include a pre-underwriting evaluation analysis scene and a post-underwriting service guarantee scene, in the insurance business claim-settlement type, the preset financial business scene may include a pre-claims investigation scene and a claims investigation scene, and in the asset management business type, the preset financial business scene may include a pre-investment evaluation analysis scene and a post-investment maintenance management scene.
Specifically, according to the financial service scene related to the target problem, the financial service scene identification of the target problem is determined, the financial service scene identification can be used for identifying the financial service scene, and the financial service scene identification can be in the forms of numbers, letters, character strings and the like, and the specific form can be set according to the actual situation. And matching the similarity between the financial service scene identifications of the N similar problem vectors obtained by searching in the preset financial problem vector library and the financial service scene identifications of the target problem vectors, and sorting the similarity of the similar problems successfully matched, for example, the similarity matching of the financial service scene identifications can be realized through a character string matching algorithm, the similarity of the similar problem vectors successfully matched is sorted through sorting algorithms such as selecting sorting or inserting sorting, and finally, M similar problem vectors which are most matched with the financial service scenes related to the target problem are obtained, wherein M is less than N.
S40: and according to the M similar question vectors, retrieving M question-answer vector groups from a preset question-answer database, wherein each question-answer vector group comprises a standard question vector and a standard answer vector.
Specifically, a preset question-answer database is connected through an index module of a langchain open source framework, question-answer pairs in the question-answer database are inquired according to codes of the obtained M similar question vectors, and question-answer vector groups corresponding to the M similar question vectors are output.
S50: and generating a scene question-answer prompt template according to the target question vector, the financial service scene identifier and M question-answer vector groups.
The scene question-answer prompt template refers to a background information prompt of a related field input to a model when a large language model is used to answer knowledge of a specific field. The scene question-answer prompting template comprises a target question vector, a financial service scene identifier and M question-answer pair vector group information, and financial service background information in the insurance field is provided for the LLM model according to the scene question-answer prompting template.
Specifically, a specific scene question-answer prompt template can be generated according to different application scenes and requirements by calling a prompt template tool or a construction function in a langchain open source framework, and only the content of the corresponding scene question-answer prompt template is required to be input.
S60: and inputting the scene question-answering prompt template into the LLM model which is trained in the question-answering of the financial business, and outputting target answer text information.
LLM model refers to a large language model based on natural language processing technology, which is trained through a large corpus, has learned semantic information and statistical features of the language, and is capable of generating consistent and meaningful answers related to a given input context.
Specifically, firstly, the generated scene question-answer prompting template is input into the LLM model through a model interaction component in a langchain open source framework, and secondly, if the input scene question-answer prompting template is text content, the text content of the scene question-answer prompting template can be converted into vector representation through a word embedding technology, and the semantic and context information provided by the scene question-answer prompting template can be obtained through the vector representation. In the invention, the input scene question-answer prompting template is vector expression, the process of converting the text into vector expression can be omitted, language knowledge and statistical characteristics obtained by pre-training are used by the LLM model, and a plurality of candidate answers are generated by combining the context information, grammar rules, vocabulary collocation and other information provided by the scene question-answer prompting template, finally, the LLM model evaluates and sorts the generated plurality of candidate answers according to the technologies of semantic matching, logical reasoning and the like, and the most suitable answer is selected and output to a user.
In this embodiment, after acquiring a financial customer service session request, the server converts a target problem into a target problem vector based on the financial customer service session request, searches N similar problem vectors in a financial problem vector library, then selects M similar problem vectors matched with a preset financial service scene identifier, searches M question-answer vector groups in a question-answer database according to the M similar problem vectors, generates a scene question-answer prompt template including the target problem vector, the financial service scene identifier and the M question-answer vector groups, and inputs the scene question-answer prompt template into a trained LLM model to output target answer text information. Compared with the existing financial insurance intelligent customer service system, the intelligent customer service system takes the scene question-answer prompt template containing insurance background information as the input of the LLM model, the LLM model improves the accuracy of target answer information by understanding and learning the known question-answer in the scene question-answer prompt template to prompt information, and on the basis, the LLM model carries out reasoning and expansion on the learned question-answer to prompt information by combining language knowledge obtained during model pre-training, generates a plurality of candidate answers, selects the most suitable answer and returns the most suitable answer to a user, improves the comprehensiveness of the target answer information, and the target answer text information and the target question generated by the LLM model can form a new question-answer pair, so that the expansion of the question-answer pair information in a database is realized, and the cost of manpower maintenance question-answer pair database is effectively reduced.
In one embodiment, as shown in fig. 3, in step S20, N similar problem vectors, N >2, are retrieved from a preset financial problem vector library according to the target problem vector, and the method includes the following steps:
s21: and extracting keywords of the target problem vector, and determining the financial business type related to the target problem according to the keywords.
The invention mainly relates to the financial business types including insurance exhibition industry, underwriting business, insurance claim settlement and asset management. Specifically, keywords of the target question vector may be extracted through a text classification algorithm, for example, target question data is classified into different categories by means of word frequency statistics and part of speech tagging on contents of the target question, keywords of each category are extracted, and financial service types related to the target question are determined according to the extracted keywords.
S22: and extracting all to-be-compared problem vectors belonging to the financial business type from the financial problem vector library.
Specifically, the to-be-compared problem vector of the same type as the financial service related to the target problem in the vector library can be extracted by using a vector index of a key-value structure, wherein the key is an index, and can be a keyword related to the target problem, the value is the to-be-compared problem vector corresponding to the keyword, and the corresponding to-be-compared problem vector is searched by the keyword related to the target problem, so that the to-be-compared vector of the financial service type related to the target problem can be obtained from the vector library.
S23: and calculating the similarity between the target problem vector and each problem vector to be compared.
Specifically, the similarity between the target problem vector and the problem vector to be compared may be calculated by a cosine similarity or euclidean distance method. For example, the similarity between the target problem vector and the problem vector to be compared is calculated through cosine similarity, the extracted problem vector to be compared and the target problem vector are expressed in the form of the same dimension matrix, and the formula of the cosine similarity is used for: similarity = vector a-vector B/(modulo length of vector a. Modulo length of vector B) to calculate the similarity between each of the to-be-compared vectors and the target problem vector.
S24: and selecting N problem vectors to be compared according to the similarity as N similar problem vectors.
And setting a similarity measurement standard among vectors, calculating the similarity between each to-be-compared problem vector and the target problem, and then carrying out sorting treatment according to the obtained similarity, and selecting N to-be-compared problem vectors reaching the similarity measurement standard as N similar problem vectors.
In this embodiment, by acquiring a keyword of a target problem, extracting to-be-compared problem vectors belonging to the same financial business type as the keyword from a financial problem vector library according to the keyword, calculating the similarity between each extracted to-be-compared problem vector and the target problem vector, and selecting N similar problem vectors according to the similarity. Through the steps, the similar problem vector with the same service type as the target problem vector is obtained, the similar problem vector is primarily screened, and the accuracy of LLM model input information is improved.
In an embodiment, as shown in fig. 4, in step S30, N similar problem vectors are ordered according to a preset financial service scene identifier, and M similar problem vectors matched with the financial service scene identifier are selected, where M < N, and the method includes the following steps:
s31: and determining a financial service scene related to the target problem according to the keywords of the target problem vector, wherein the financial service scene comprises a pre-sale consultation scene and an after-sale service scene.
In the insurance exhibition business type, the financial business scene comprises a consultation scene before insurance sales and a service scene after insurance sales, and the financial business scene also comprises a financial business scene in the underwriting business type, the insurance business claim settlement type and the asset management business type, and the specific financial business scene division can be adjusted according to actual conditions.
Specifically, according to the keywords of the target problem vector, determining a financial service scene related to the target problem, wherein the financial service scene is formed by further dividing financial service types related to the target problem into a pre-sale consultation scene and an after-sale service scene.
S32: and determining the identification of the financial business scene of the target problem according to the financial business scene related to the target problem.
The financial service scene identifier can be used for indicating a financial service scene, and can be in the forms of numbers, letters, character strings and the like, and the specific form can be set according to actual conditions.
S33: and matching the financial service scene identification of the target problem with the financial service scene identifications of N similar problem vectors, and filtering the similar problem vectors which are not successfully matched.
Specifically, the financial service scene identifier of each similar problem vector and the financial service scene identifier of the target problem can be respectively matched through a character string matching algorithm, for example, the financial service scene identifier of each similar problem vector and the financial service scene identifier of the target problem are matched through a violent matching algorithm in the character string matching algorithm, the financial service scene identifier of the target problem is set as a mode string, the financial service scene identifier of each similar problem vector is set as a target character string, the first character of each target character string is sequentially started, the financial service scene identifiers are compared with the mode string one by one, if the financial service scene identifiers are not matched, the similar problem vector and the target problem vector are considered to belong to different financial service scene identifiers until N financial service scene identifiers of the similar problem vectors are completely matched, and the similar problem vectors which are unsuccessfully matched are filtered.
S34: and sorting the similar problem vectors successfully matched according to the similarity between the target problem vector and each similar problem vector, and selecting M similar problem vectors.
Specifically, the similarity magnitude between each similar problem vector and the target problem vector may be ranked by selecting a ranking algorithm. Traversing the whole similarity sequence to be sequenced from the similarity between the first similarity problem vector and the target problem vector, assuming the current similarity to be the minimum value, marking the similarity as the minimum value index, comparing the similarity with the minimum similarity from the next position of the current similarity, updating the minimum value index until traversing the whole similarity sequence to be sequenced, finding the minimum similarity, and repeating the steps, and sequentially determining the minimum value in the unordered similarity until sequencing all the similarities. And selecting M similar problem vectors according to the ordered similar problem vectors.
In this embodiment, a financial service scenario related to a target problem is determined, similarity matching is performed according to the service scenario and N similar problem vectors extracted from a financial problem vector library, so as to filter out similar problem vectors which are not successfully matched, the similar problem vectors which are successfully matched are ordered according to the similarity between the matching similar problem vectors and the target problem, and finally M similar problem vectors which are most consistent with the user problem are selected. The N similar problem vectors extracted from the financial problem vector library are further filtered, matched and sequenced in similarity according to the financial business scene, M similar problem vectors which are most in line with the target problem are selected, the similar problem vectors are further screened, clear and definite subject range and background information are provided for the input of the LLM model, and the accuracy of target answer information is further improved.
In one embodiment, in step S50, generating a scene question-answer prompt template according to the target question vector, the financial service scene identifier, and M question-answer vector groups, including:
and sending the target problem vector, the financial service scene identifier and M question-answering vector groups to a prompt template component in a langchain open source framework core module, and generating the scene question-answering prompt template by using the prompt template component.
Specifically, a specific scene question-answer prompt template can be generated according to different application scenes and requirements by calling a prompt template tool or a constructor in a langchain open source framework, wherein the specific form of the scene question-answer prompt template can be designed according to specific scenes.
In this embodiment, by generating a scene question-answer prompt template including a target question vector, a financial service scene identifier and M question-answer vector groups, clear and comprehensive background information is provided for the input of the LLM model, and the LLM model can limit the answer range of the target question in a specific service scene of a specific service type in the financial insurance field according to the scene question-answer prompt template, thereby improving the accuracy of the answer information of the LLM model.
In one embodiment, in step S50, the generated scene question-answer prompting template includes:
the system comprises a known prompt information identifier, a known prompt information content, a problem prompt information identifier and a problem prompt content, wherein the known prompt information content comprises M question-answer vector groups and the financial service scene identifiers, and the problem prompt content comprises the target problem vector.
Specifically, the known prompt information identifier is used for identifying the known prompt information content, and the problem prompt information identifier is used for identifying the problem prompt content. The known prompt information content and the problem prompt content can be expressed in a text or vector form, and the purpose of the known prompt information content and the problem prompt content is to provide comprehensive and clear background information for the LLM model so that the user can better understand the requirements of the user and generate comprehensive and accurate answers based on the comprehensive and clear background information.
The content of the scene question-answer prompting template can be a simple question, an instruction or some examples and the like, and can be adjusted according to practical situations, but the scene question-answer prompting template needs to comprise specific questions or instructions, detailed background information or other information which is helpful for solving the questions and proper grammar and format.
For example, the scene question and answer prompt template may be:
prompt information is known:
consultation scene before sale of policy
{ question-answer pair 1: which insurance types can be purchased over 50?
The 50 th old people can buy accident risks, serious disease risks, commercial medical risks, endowment risks, cancer prevention risks, endowment risks and life risks.
Question-answer pair 2: what does a parent not pay a worker's medical insurance to get retired money when he wants to retire?
If the old people do not pay staff medical insurance in the past, people over 50 years old can choose to pay urban and rural resident endowment insurance, and can pay for fifteen years at a time, and retirement can be taken every month.
Question and answer pair 3: what are medical risks of low premium and high premium?
If the risk of the small illness is to be dealt with, the small medical insurance can be purchased, the premium is cheap, the reimbursement is low, and part of the small medical insurance is even free of reimbursement. If the patient wants to deal with the risk of serious illness, the patient can buy millions of medical insurance, the premium is cheap and high, and the claim free amount is relatively high. }
Questions are answered concisely and professionally based on the known information. If the answer can not be obtained from the question, please say "very sorry, the question can not be answered according to the medical information, and please contact with the manual customer service. "or" you do not provide enough information about the problem, please re-describe the problem or contact manual service in detail. "do not allow the addition of a composition to the answer, the answer is in Chinese.
Question prompting information:
{ buy insurance for father who has not paid employee's medical insurance around 50 years old, want to buy inexpensive guaranteed insurance, should what should be chosen? }
In this embodiment, the contents of the scene question-answer prompt template include a known prompt information identifier, a known prompt information content, a problem prompt information identifier, and a problem prompt information content, where the known prompt information content includes M question-answer vector groups and a financial service scene identifier, and the problem prompt content includes a target problem vector. The known prompt information content provides a related question-answer pair background prompt for the input of the LLM model, the target question vector confirms the subject of the question and the answer and operation expected by the user, and the LLM model can better understand the intention of the user according to the scene question-answer prompt template, so that the comprehensiveness and accuracy of the target answer text information are improved.
In one embodiment, as shown in fig. 5, in step S60, the scene question-answer prompt template is input to the LLM model which has been trained for the financial service question-answer, and the target answer text information is output, including the following steps:
s61: and sending the scene question-answer prompt template to a model interaction component in a langchain open source framework core module, inputting the scene question-answer prompt template to the LLM model which is trained in the financial business question-answer by using the model interaction component, and outputting the target answer text information through the LLM model.
Specifically, the scene question-answer prompt template is transferred to the LLM model as parameters by calling an interface provided by the langchain open source framework, after the scene question-answer prompt template is input, the content of the scene question-answer prompt template is coded and processed through a pre-trained language model or word embedding technology, the content is converted into a vector form which can be understood and processed by the LLM model, the LLM model generates a plurality of answers through understanding and analyzing the scene question-answer prompt template, and the most suitable answers are converted into texts through a decoder and are output to a user.
S62: and converting the target answer text information into target answer voice information, and outputting the target answer voice information or the target answer text information.
Specifically, according to the actual application scenario, the target answer text information can be converted into target voice information through a voice synthesis technology, and the target answer text information mainly comprises two stages of text analysis and voice synthesis. In the text analysis stage, preliminary word segmentation, grammar analysis and other preprocessing operations are performed on the target answer text information, in the speech synthesis stage, firstly, the input target answer text information is processed through an acoustic model to generate corresponding speech synthesis parameters, the speech synthesis parameters comprise fundamental frequency, formant frequency and the like, the characteristics of sound can be described, then, the speech synthesis parameters are converted into corresponding speech waveforms through a signal processing technology, for example, the signal processing technology can be a rule-based synthesis method, a deep learning-based generation model and the like, finally, post-processing operations such as sound smoothing, denoising and the like are performed on the generated speech waveforms, and the processed speech information is output to a user.
In this embodiment, under the langchain open source framework, the scene question-answer prompt template is input into the LLM model which has been trained for answering financial services, and the corresponding target answer text information is output through LLM model processing. By inputting the background information of the scene question-answer prompt template into the LLM model, the LLM model can provide more comprehensive and accurate answers for users based on specific business scene subjects in the specific financial insurance field, and the answers can be in the form of voice or text, so that the convenience of the users for carrying out financial insurance consultation services is further improved.
In one embodiment, as shown in fig. 6, in step S40, M question-answer vector groups are retrieved from a preset question-answer database according to M similar question vectors, and each question-answer vector group includes a standard question vector and a standard answer vector, including the following steps:
s41: and retrieving M question-answer vector groups from a preset question-answer database by utilizing an index module component in a langchain open-source framework core module through interaction between the preset first chain component and the question-answer database.
Chain components in the langchain open source framework refer to components of one data path that combine multiple components between models or systems. The indexing module is a unique identifier for identifying language resources, through which specific database resources can be quickly retrieved and accessed.
Specifically, the first chain component links the index module and the question-answer database, can be connected to the question-answer pair database through the first chain component, queries question-answer pairs in the question-answer pair database according to the obtained codes of M similar question vectors, and outputs question-answer vector groups corresponding to the M similar question vectors.
S42: and sending the question-answer pair vector group to the prompt template component by utilizing the question-answer pair database through interaction connection between a preset second chain component and the prompt template component.
The second chain component links the question-answer pair database and the prompt template component, through which the question-answer pair vector group can be sent to the prompt template component.
In this embodiment, the index module and the question-answer database are connected through the first chain component, M question-answer vector groups are retrieved from the preset question-answer database, the question-answer database and the prompt template component are connected through the second chain component, and the question-answer vector groups are sent to the prompt template component. M question-answering vector groups corresponding to M similar questions are sent to a prompt template component through a first chain component and a second chain component in a langchain open source core framework, so that a convenient and consistent data path is provided for generating a scene question-answering prompt template, and the running speed of a program is improved.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present invention.
In an embodiment, a financial customer service session generating device based on a LLM model is provided, where the financial customer service session generating device based on the LLM model corresponds to the financial customer service session generating method based on the LLM model in the above embodiment one by one. As shown in fig. 7, the LLM model-based financial customer service dialogue generating device includes a target question converting module 71, a similar question retrieving module 72, a similar question ordering module 73, a question-answer pair retrieving module 74, a question-answer prompt template generating module 75, and a model outputting module 76. The functional modules are described in detail as follows:
a target question conversion module 71, configured to obtain a target question, and convert the target question into a target question vector;
the similar problem retrieving module 72 is configured to retrieve N similar problem vectors, N >2, from a preset financial problem vector library according to the target problem vector;
the similarity problem ordering module 73 is configured to order N similarity problem vectors according to a preset financial service scene identifier, and select M similarity problem vectors that are matched with the financial service scene identifier, where M < N;
A question-answer pair retrieval module 74, configured to retrieve M question-answer pair vector groups from a preset question-answer database according to M similar question vectors, where each question-answer pair vector group includes a standard question vector and a standard answer vector;
a question-answer prompting template generating module 75, configured to generate a scene question-answer prompting template according to the target question vector, the financial service scene identifier, and M question-answer vector groups;
the model output module 76 is configured to input the scenario question-answer prompt template into the LLM model that has been trained for the financial service question-answer, and output the target answer text information.
Optionally, the similar problem retrieval module 72 further includes:
the business type determining module is used for extracting keywords of the target problem vector and determining the financial business type related to the target problem according to the keywords;
the to-be-compared problem extraction module is used for extracting all to-be-compared problem vectors belonging to the financial business type in the financial problem vector library;
the similarity calculation module is used for calculating the similarity between the target problem vector and each problem vector to be compared;
And the similarity problem selection module is used for selecting N problem vectors to be compared according to the similarity as N similar problem vectors.
Optionally, the similar problem ordering module 73 further includes:
the business scene determining module is used for determining a financial business scene related to the target problem according to the keyword of the target problem vector, wherein the financial business scene comprises a pre-sale consultation scene and an after-sale service scene of a policy;
the business scene identification determining module is used for determining the financial business scene identification of the target problem according to the financial business scene related to the target problem;
the matching module is used for matching the financial service scene identifications of the target problems with the financial service scene identifications of N similar problem vectors and filtering the similar problem vectors which are not successfully matched;
and the similar problem ordering module is used for matching the financial service scene identifiers of the target problems with the financial service scene identifiers of N similar problem vectors and filtering the similar problem vectors which are not successfully matched.
Optionally, the question-answer prompting template generating module 75 further includes:
and the sending module is used for sending the target problem vector, the financial service scene identifier and M question-answer vector groups to a prompt template component in the langchain open source framework core module, and generating the scene question-answer prompt template by using the prompt template component.
Optionally, the question-answer prompting template generating module 75 further includes:
the scene question-answering prompt module is used for storing known prompt information identifiers, known prompt information contents, problem prompt information identifiers and problem prompt contents, wherein the known prompt information contents comprise M question-answering vector groups and the financial service scene identifiers, and the problem prompt contents comprise the target problem vectors.
Optionally, the model output module 76 further includes:
the model input/output module is used for sending the scene question-answer prompt template to a model interaction component in a langchain open source framework core module, inputting the scene question-answer prompt template to the LLM model which is subjected to the financial service question-answer training by utilizing the model interaction component, and outputting the target answer text information through the LLM model;
and the target answer conversion module is used for converting the target answer text information into target answer voice information and outputting the target answer voice information or the target answer text information.
Optionally, the question-answer pair search module 74 further includes:
the first chain component connection module is used for searching M question-answer vector groups in a preset question-answer database through the interaction between the preset first chain component and the question-answer database by utilizing an index module component in the langchain open source framework core module;
And the second chain component connection module is used for sending the question-answer pair vector group to the prompt template component through the interaction connection between the preset second chain component and the prompt template component by utilizing the question-answer pair database.
For specific limitations on the LLM model-based financial customer service session generation apparatus, reference may be made to the above limitations on the LLM model-based financial customer service session generation method, and no further description is given here. The modules in the financial customer service session generating device based on the LLM model may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
Fig. 8 is a schematic structural diagram of a terminal device according to a fourth embodiment of the present invention. As shown in fig. 8, the terminal device of this embodiment includes: at least one processor (only one shown in fig. 8), a memory, and a computer program stored in the memory and executable on the at least one processor, the processor, when executing the computer program, implementing the steps in the embodiments of the LLM model-based financial customer service session generation method described above.
The terminal device may include, but is not limited to, a processor, a memory. It will be appreciated by those skilled in the art that fig. 8 is merely an example of a terminal device and is not limiting of the terminal device, and that the terminal device may comprise more or less components than shown, or may combine some components, or different components, e.g. may further comprise a network interface, a display screen, an input device, etc.
The processor may be a CPU, but may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory includes a readable storage medium, an internal memory, etc., where the internal memory may be a memory of the terminal device, and the internal memory provides an environment for the operation of an operating system and computer readable instructions in the readable storage medium. The readable storage medium may be a hard disk of the terminal device, and in other embodiments may be an external storage device of the terminal device, for example, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card), etc. that are provided on the terminal device. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used to store an operating system, application programs, boot loader (BootLoader), data, and other programs such as program codes of computer programs, and the like. The memory may also be used to temporarily store data that has been output or is to be output.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment 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, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present invention. The specific working process of the units and modules in the above device may refer to the corresponding process in the foregoing method embodiment, which is not described herein again. 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 present invention may implement all or part of the flow of the method of the above-described embodiment, and may be implemented by a computer program to instruct related hardware, and the computer program may be stored in a computer readable storage medium, where the computer program, when executed by a processor, may implement the steps of the method embodiment described above. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, executable files or in some intermediate form, etc. The computer readable medium may include at least: any entity or device capable of carrying computer program code, a recording medium, a computer Memory, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium. Such as a U-disk, removable hard disk, magnetic or optical disk, etc. In some jurisdictions, computer readable media may not be electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
The present invention may also be implemented by a computer program product for implementing all or part of the steps of the method embodiments described above, when the computer program product is run on a terminal device, causing the terminal device to execute the steps of the method embodiments described above.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided in the present invention, it should be understood that the disclosed apparatus/terminal device and method may be implemented in other manners. For example, the apparatus/terminal device embodiments described above are merely illustrative, e.g., the division of modules or 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 may be an indirect coupling or communication connection via interfaces, devices or units, which may be in electrical, mechanical or other forms.
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 over 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.
The above embodiments are only for illustrating the technical aspects of the present invention, not for limiting the same, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may be modified or some technical features may be replaced with other technical solutions, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and all the modifications or replacements are included in the protection scope of the present invention.
Claims (10)
1. The financial customer service dialogue generation method based on the LLM model is characterized by comprising the following steps of:
acquiring a target problem, and converting the target problem into a target problem vector;
According to the target problem vector, N similar problem vectors are obtained through searching in a preset financial problem vector library, wherein N is more than 2;
according to a preset financial service scene identifier, sorting N similar problem vectors, and selecting to obtain M similar problem vectors matched with the financial service scene identifier, wherein M is less than N;
according to the M similar question vectors, retrieving M question-answer vector groups from a preset question-answer database, wherein each question-answer vector group comprises a standard question vector and a standard answer vector;
generating a scene question-answer prompt template according to the target question vector, the financial service scene identifier and M question-answer vector groups;
and inputting the scene question-answering prompt template into the LLM model which is trained in the question-answering of the financial business, and outputting target answer text information.
2. The LLM model-based financial customer service dialogue generation method as set forth in claim 1, wherein the retrieving N similar problem vectors, N >2, from a preset financial problem vector library according to the target problem vector comprises:
extracting keywords of the target problem vector, and determining the financial service type related to the target problem according to the keywords;
Extracting all to-be-compared problem vectors belonging to the financial business type from the financial problem vector library;
calculating the similarity between the target problem vector and each problem vector to be compared;
and selecting N problem vectors to be compared according to the similarity as N similar problem vectors.
3. The method for generating a financial customer service dialogue based on LLM model as claimed in claim 1, wherein said sorting N similar problem vectors according to a preset financial service scene identifier, selecting M similar problem vectors matching with the financial service scene identifier, M < N, includes:
determining a financial service scene related to the target problem according to the keywords of the target problem vector, wherein the financial service scene comprises a policy after-sale scene and a policy before-sale consultation scene;
determining a financial service scene identification of the target problem according to the financial service scene related to the target problem;
matching the financial service scene identification of the target problem with the financial service scene identifications of N similar problem vectors, and filtering similar problem vectors which are not successfully matched;
And sorting the similar problem vectors successfully matched according to the similarity between the target problem vector and each similar problem vector, and selecting M similar problem vectors.
4. The LLM model-based financial customer service dialogue generation method as set forth in claim 1, wherein the generating a scene question-answer prompt template from the target question vector, the financial service scene identification, and M question-answer vector groups comprises:
and sending the target problem vector, the financial service scene identifier and M question-answering vector groups to a prompt template component in a langchain open source framework core module, and generating the scene question-answering prompt template by using the prompt template component.
5. The LLM model-based financial customer service dialogue generation method as claimed in claim 1, wherein the scene question-answer prompt template comprises:
the system comprises a known prompt information identifier, a known prompt information content, a problem prompt information identifier and a problem prompt content, wherein the known prompt information content comprises M question-answer vector groups and the financial service scene identifiers, and the problem prompt content comprises the target problem vector.
6. The LLM model-based financial customer service dialogue generation method as set forth in claim 1, wherein the inputting the scene question-answer prompt template into the LLM model having undergone the training of the financial service question-answer, outputting the target answer text information, comprises:
the scene question-answer prompt template is sent to a model interaction component in a langchain open source framework core module, the model interaction component is utilized to input the scene question-answer prompt template to the LLM model which is subjected to the financial business question-answer training, and the target answer text information is output through the LLM model;
and converting the target answer text information into target answer voice information, and outputting the target answer voice information or the target answer text information.
7. The LLM model-based financial customer service dialogue generation method as set forth in claim 1, wherein M question-answer vector sets are retrieved from a preset question-answer database according to M similar question vectors, each of the question-answer vector sets including a standard question vector and a standard answer vector, comprising:
an index module component in a langchain open source framework core module is utilized to search M question-answer vector groups in a preset question-answer database through the interaction between the preset first chain component and the question-answer database;
And sending the question-answer pair vector group to the prompt template component by utilizing the question-answer pair database through interaction connection between a preset second chain component and the prompt template component.
8. A financial customer service session generation device based on LLM model, comprising:
the target problem conversion module is used for acquiring a target problem and converting the target problem into a target problem vector;
the similar problem retrieval module is used for retrieving N similar problem vectors in a preset financial problem vector library according to the target problem vector, wherein N is more than 2;
the similar problem ordering module is used for ordering the N similar problem vectors according to a preset financial service scene identifier, and selecting M similar problem vectors matched with the financial service scene identifier, wherein M is smaller than N;
the question-answer pair retrieval module is used for retrieving M question-answer pair vector groups from a preset question-answer database according to the M similar question vectors, wherein each question-answer pair vector group comprises a standard question vector and a standard answer vector;
the question-answer prompting template generation module is used for generating a scene question-answer prompting template according to the target question vector, the financial service scene identifier and M question-answer vector groups;
And the model output module is used for inputting the scene question-answer prompt template into the LLM model which is trained by the financial business question-answer, and outputting target answer text information.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the LLM model-based financial customer service session generation method as claimed in any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the LLM model-based financial customer service session generation method as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311494476.0A CN117520503A (en) | 2023-11-09 | 2023-11-09 | Financial customer service dialogue generation method, device, equipment and medium based on LLM model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311494476.0A CN117520503A (en) | 2023-11-09 | 2023-11-09 | Financial customer service dialogue generation method, device, equipment and medium based on LLM model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117520503A true CN117520503A (en) | 2024-02-06 |
Family
ID=89754391
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311494476.0A Pending CN117520503A (en) | 2023-11-09 | 2023-11-09 | Financial customer service dialogue generation method, device, equipment and medium based on LLM model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117520503A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117744804A (en) * | 2024-02-19 | 2024-03-22 | 粤港澳大湾区数字经济研究院(福田) | Reasoning method, terminal and medium of financial analysis task based on large language model |
-
2023
- 2023-11-09 CN CN202311494476.0A patent/CN117520503A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117744804A (en) * | 2024-02-19 | 2024-03-22 | 粤港澳大湾区数字经济研究院(福田) | Reasoning method, terminal and medium of financial analysis task based on large language model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112346567B (en) | Virtual interaction model generation method and device based on AI (Artificial Intelligence) and computer equipment | |
| CN112818093B (en) | Evidence document retrieval method, system and storage medium based on semantic matching | |
| CN111368042A (en) | Intelligent question and answer method and device, computer equipment and computer storage medium | |
| CN112287069B (en) | Information retrieval method and device based on voice semantics and computer equipment | |
| CN110390084A (en) | Text duplicate checking method, apparatus, equipment and storage medium | |
| CN111078837A (en) | Intelligent question and answer information processing method, electronic equipment and computer readable storage medium | |
| CN111782793A (en) | Intelligent customer service processing method, system and equipment | |
| CN117520503A (en) | Financial customer service dialogue generation method, device, equipment and medium based on LLM model | |
| CN115687647A (en) | Notarization document generation method and device, electronic equipment and storage medium | |
| CN112507095A (en) | Information identification method based on weak supervised learning and related equipment | |
| CN114398473A (en) | Enterprise portrait generation method, device, server and storage medium | |
| CN113821612A (en) | Information searching method and device | |
| CN116402166B (en) | Training method and device of prediction model, electronic equipment and storage medium | |
| CN113033912A (en) | Problem solving person recommendation method and device | |
| CN112199958A (en) | Concept word sequence generation method and device, computer equipment and storage medium | |
| CN115358817A (en) | Intelligent product recommendation method, device, equipment and medium based on social data | |
| CN115221323A (en) | Cold start processing method, device, equipment and medium based on intention recognition model | |
| CN114138954A (en) | User consultation problem recommendation method, system, computer equipment and storage medium | |
| CN114067343A (en) | Data set construction method, model training method and corresponding device | |
| CN114117057A (en) | Keyword extraction method of product feedback information and terminal equipment | |
| CN111382246A (en) | Text matching method, matching device and terminal | |
| CN117932009A (en) | ChatGLM model-based insurance customer service dialogue generation method, chatGLM model-based insurance customer service dialogue generation device, chatGLM model-based insurance customer service dialogue generation equipment and ChatGLM model-based insurance customer service dialogue generation medium | |
| CN116431779B (en) | FAQ question-answering matching method and device in legal field, storage medium and electronic device | |
| CN112597289B (en) | Query information classification method, model training method, device and electronic equipment | |
| CN114723523B (en) | Product recommendation method, device, equipment and medium based on user capability image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |