Recording information generation method and device, electronic equipment and computer program product

Abstract

The invention discloses a notice information generation method and device, electronic equipment and a computer program product. The method comprises the following steps: acquiring collection call recording data and converting the collection call recording data into collection text data; according to the collection text data and first preset prompt information, utilizing a notice information generation model to generate notice summary information, the notice information generation model being obtained by fine tuning a pre-trained large language model; and according to the notation reminding summary information and second preset prompt information, generating notation reminding result form information by utilizing a notation reminding information generation model. According to the method and the device, the announcement information is automatically generated by adopting the announcement information generation model, and the announcement information generation task is split into the announcement summary task and the structured announcement result information extraction task by utilizing the text summary generation capability of a large language model; and the structured collection result form data is further extracted by using the summary text containing the key information, so that the generation efficiency of the reminder information is improved, and the working efficiency and accuracy of collection personnel are further improved.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a method, device, electronic device, and computer program product for generating reminder information. Background Art

[0002] As an important indicator to measure the quality of bank assets, the non-performing asset ratio reflects the proportion of overdue and non-performing loans in the bank's loan portfolio. Post-loan management is particularly important as the main means to reduce the non-performing asset ratio. Collection is a key link in post-loan management. Calling and communicating with overdue customers is the most effective and main way. How to effectively use appropriate words to collect debts, improve collection efficiency, and encourage customers to fulfill their obligations and repay on time has become an issue that banks need to continuously optimize.

[0003] The existing collection methods mainly rely on manual filling by the collection staff, which is inefficient and difficult to guarantee accuracy. There are also some methods that rely on pattern matching and machine learning models to establish a mapping between call data and collection information. Only one hierarchical model is used to generate the repayment willingness label. On the one hand, the amount of information obtained is too little, and on the other hand, the generalization is poor and the accuracy is low. Summary of the invention

[0004] The embodiments of the present disclosure provide a method, device, electronic device, and computer program product for generating reminder information to improve the generation efficiency and accuracy of reminder information and improve the collection efficiency.

[0005] The embodiments of the present disclosure adopt the following technical solutions:

[0006] In a first aspect, an embodiment of the present disclosure provides a method for generating reminder information, the method comprising:

[0007] Obtain collection call recording data;

[0008] Converting the collection call recording data into collection text data;

[0009] Generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model;

[0010] According to the reminder summary information and the second preset prompt information, the reminder information generation model is used to generate reminder result form information.

[0011] Optionally, the generating the reminder summary information by using the reminder information generation model according to the reminder text data and the first preset prompt information includes:

[0012] Constructing a first preset prompt word corresponding to the reminder summary information;

[0013] The reminder summary information is generated by using the reminder information generation model according to the reminder text data and the first preset prompt word.

[0014] Optionally, the generating reminder result form information by using the reminder information generation model according to the reminder summary information and the second preset prompt information includes:

[0015] Constructing a second preset prompt word corresponding to the reminder result form information;

[0016] Obtain the corresponding generation example of reminder summary information and reminder result form information;

[0017] The reminder result form information is generated using the reminder information generation model according to the reminder summary information, the second preset prompt word corresponding to the reminder result form information, and the corresponding generation example of the reminder summary information and the reminder result form information.

[0018] Optionally, the reminder result form information is structured reminder result form information, and the structured reminder result form information includes at least one of a customer communication result label, a reason for overdue payment, an expected repayment time, and a customer attitude. The generation of the reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information includes:

[0019] Fill in a reminder result form based on the structured reminder result form information.

[0020] Optionally, the reminder information generation model is trained in the following manner:

[0021] Constructing collection reminder information to generate sample data, wherein the training sample data is constructed based on historical collection call recording data and corresponding historical collection reminder information;

[0022] A pre-trained large language model is obtained, and the pre-trained large language model is fine-tuned using the training sample data to obtain a trained reminder information generation model.

[0023] Optionally, the historical reminder information includes historical reminder summary information, and the step of constructing reminder information to generate sample data includes:

[0024] Obtain historical collection call recording data and corresponding historical collection summary information;

[0025] Converting the historical debt collection call recording data into historical debt collection text data;

[0026] The reminder information generation sample data is constructed according to the historical reminder text data and the corresponding historical reminder summary information.

[0027] Optionally, the historical reminder information further includes historical reminder result form information, and obtaining the corresponding historical reminder summary information includes:

[0028] Determine whether the historical collection call recording data has corresponding historical collection summary information;

[0029] If it exists, directly obtain the historical collection summary information corresponding to the call recording data of the historical collection;

[0030] Otherwise, the corresponding historical reminder summary information is generated according to the historical reminder result form information.

[0031] In a second aspect, the embodiment of the present disclosure further provides a reminder information generation device, characterized in that the reminder information generation device comprises:

[0032] An acquisition unit, used to acquire collection call recording data;

[0033] A conversion unit, used to convert the collection call recording data into collection text data;

[0034] A first generating unit, configured to generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model;

[0035] The second generating unit is used to generate reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information.

[0036] In a third aspect, an embodiment of the present disclosure further provides an electronic device, including:

[0037] A processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes any of the aforementioned reminder information generation methods.

[0038] In a fourth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements any of the aforementioned reminder information generation methods.

[0039] At least one of the above technical solutions adopted in the embodiments of the present disclosure can achieve the following beneficial effects: the method for generating reminder information in the embodiments of the present disclosure first obtains the collection call recording data; then converts the collection call recording data into collection text data; then generates reminder summary information based on the collection text data and the first preset prompt information using the reminder information generation model, the reminder information generation model is obtained by fine-tuning the pre-trained large language model; finally, generates reminder result form information based on the reminder summary information and the second preset prompt information using the reminder information generation model. The method for generating reminder information in the embodiments of the present disclosure uses the reminder information generation model obtained by fine-tuning the pre-trained large language model to automatically generate reminder information, and utilizes the text summary generation capability that the large language model is good at, splits the reminder information generation task into two tasks: reminder summary and structured reminder result information extraction, and uses the summary text containing key information to further extract structured collection result form data, thereby improving the efficiency of reminder information generation, and thus improving the work efficiency and accuracy of the collection personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0041] Figure 1 A schematic diagram of a process for generating reminder information in an embodiment of the present disclosure;

[0042] Figure 2 A schematic diagram of the overall process of generating reminder information in an embodiment of the present disclosure;

[0043] Figure 3 It is a structural schematic diagram of a reminder information generating device in an embodiment of the present disclosure;

[0044] Figure 4 It is a structural schematic diagram of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0046] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0047] The main technical terms involved in this disclosure include:

[0048] (1) Large language models: Usually based on neural network architectures, such as the Transformer model, which uses a self-attention mechanism to understand the complex relationships between words and sentences in text. The model gains a deep understanding of language by learning from a large amount of text data, including books, articles, web page content, etc.

[0049] The large number of parameters in a large language model means that it can capture more language features and patterns, and thus perform better when handling complex language tasks. The model is first trained in an unsupervised pre-training phase to learn the general characteristics of the language, and is fine-tuned on various specific, supervised tasks to adapt to specific application scenarios. However, large language models also have some limitations and challenges, such as large computing resource requirements, high training costs, and the possibility of inaccurate or biased content. Therefore, when using large language models, it is necessary to carefully consider the issue of bias and take appropriate measures to ensure the accuracy and fairness of their output.

[0050] The application scenarios of large language models are very wide. For example, in the field of text generation, large language models can be used for automatic writing, generating news reports, writing advertising copy, etc. In the field of machine translation, large language models can translate one language into another, such as translating Chinese into English. In the field of text summarization, large language models can generate summaries of articles to help readers quickly understand the main content of the article. In the field of question answering, large language models can answer users' questions, such as intelligent assistants, customer service robots, etc.

[0051] (2) ChatGLM2-6B: This is a conversation pre-training model jointly released by Zhipu AI and Tsinghua University KEG Laboratory. It has about 6 billion parameters and is an open source model in the ChatGLM2 series. While retaining many excellent features of the first-generation model, such as smooth conversation and low deployment threshold, it has the following new features:

[0052] 1) Longer context: Based on FlashAttention technology, the context length of the base model is extended from 2K in ChatGLM-6B to 32K, and 8K context length is used for training in the dialogue stage, allowing more rounds of dialogue;

[0053] 2) More efficient reasoning: Based on Multi-QueryAttention technology, ChatGLM2-6B has more efficient reasoning speed and lower video memory usage;

[0054] 3) Stronger performance: ChatGLM2-6B uses the hybrid objective function of GLM and has undergone pre-training on 1.4T Chinese and English identifiers and human preference alignment training. Evaluation results show that compared with the first-generation model, ChatGLM2-6B has achieved significant performance improvements in multiple tasks and is highly competitive among open source models of the same size.

[0055] (3) LoRA (Low-Rank Adaptation): It is a model fine-tuning method. The principle is to freeze the pre-trained model weights and inject the trainable rank decomposition matrix into each layer of the model architecture system, thereby greatly reducing the number of trainable parameters of downstream tasks. Moreover, this method takes up very little resources and is suitable for small sample scenarios. Its advantage is that it can be performed when the training data and computing resources are limited, without the need to adjust all the model parameters.

[0056] The disclosed embodiment also provides a method for generating reminder information, such as Figure 1 As shown, a flowchart of a method for generating reminder information in an embodiment of the present disclosure is provided. The method for generating reminder information at least includes the following steps S110 to S140:

[0057] Step S110, obtaining collection call recording data.

[0058] Combination Figure 2 , provides a schematic diagram of the overall process of generating reminder information in an embodiment of the present disclosure. The generation task of reminder information belongs to the natural language understanding and generation task in the field of artificial intelligence. The generation of reminder information in the present disclosure mainly includes the generation of reminder summary and reminder result form.

[0059] When generating collection reminder information, it is necessary to first obtain the real-time call recording data between the collection personnel and the collection personnel during the collection process as the basis for the subsequent generation of collection reminder information.

[0060] Step S120, converting the collection call recording data into collection text data.

[0061] In order to extract useful information from call recording data, it is necessary to convert the call recording data into text data. This step usually involves speech recognition technology. By performing speech recognition on call recording data, the speech content can be accurately converted into corresponding text content, which is convenient for subsequent processing and analysis.

[0062] Step S130, generating reminder summary information using a reminder information generation model according to the reminder collection text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model.

[0063] After obtaining the collection text content, the collection text data and the first preset prompt information are combined and input into the collection information generation model, and the collection information generation model is used to automatically generate a collection summary. The collection summary summarizes the key conclusions of the collection dialogue process, which is similar to a collection call summary.

[0064] The reminder information generation model disclosed in the present invention is obtained by fine-tuning a pre-trained large language model, and the fine-tuning method can be implemented by, for example, LoRA (Low-Rank Adaptation), etc. Compared with directly training a large language model, the manpower and time costs are greatly reduced.

[0065] The role of the first preset prompt information is to guide and drive the reminder summary model to generate reminder summary information that meets actual requirements, fully tap the capabilities of the large language model, and flexibly design prompt information according to actual needs, which is not specifically limited here.

[0066] Step S140, generating reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information.

[0067] The purpose of generating the reminder result form information is to structure the entire reminder information and extract important information from it to fill in the reminder result table data, such as customer communication result labels, overdue reasons, expected repayment time and customer attitude, etc. Therefore, the reminder summary information generated in the above steps can be used as the basis for generating the reminder result form information. In addition, considering that the reminder summary and reminder result form information have different structures or forms of expression, the two have different generation requirements for the reminder information generation model. The corresponding second preset prompt information can be designed for the generation of the reminder result form information, so as to guide and drive the model to generate reminder result form information that meets the form requirements.

[0068] By using a large language model to assist in generating the above collection information, it can effectively save the time of collection personnel and increase the number of collection cases handled daily. On the one hand, it can track the repayment status of customers in the future to see whether they have repaid according to the communication time. On the other hand, before the next collection, based on the previous collection record information and historical repayment behavior data, a more reasonable collection strategy that better matches the customer can be generated.

[0069] The collection information generation method of the disclosed embodiment adopts a collection information generation model obtained by fine-tuning a pre-trained large language model to automatically generate collection information. It utilizes the text summary generation capability of the large language model, splits the collection information generation task into two tasks: collection summary and structured collection result information extraction. The summary text containing key information is used to further extract structured collection result form data, thereby improving the collection information generation efficiency and further improving the work efficiency and accuracy of collection personnel.

[0070] In some embodiments of the present disclosure, generating reminder summary information using a reminder information generation model based on the reminder text data and the first preset prompt information includes: constructing a first preset prompt word corresponding to the reminder summary information; generating reminder summary information using the reminder information generation model based on the reminder text data and the first preset prompt word.

[0071] The generation of reminder summary information is mainly to generate a reminder summary record text. Therefore, when designing the first preset prompt information, the prompt word Prompt is mainly designed. The prompt word is a text description that drives the large model to complete the task. When the task is input to the model, natural language is used to indicate what task the model should perform or what output it should generate. High-quality prompt words can improve the generation effect of the large model and fully tap the capabilities of the pre-trained model.

[0072] The present disclosure aims at the goal of generating a collection summary, and corresponding prompt words can be designed to guide the model to generate a better response, so as to maximize the model's capabilities. For example, in the collection process, key information of concern may include the debtor's willingness to repay, repayment ability, repayment plan, etc., while specific situations may include the debtor's special circumstances (such as unemployment, illness, etc.), the collection staff's communication strategy, etc. Therefore, the prompt words used to generate the collection summary can include words or phrases related to these key information and specific situations, such as "willingness to repay", "repayment plan", "special situation explanation", etc. Of course, how to design the content of the prompt words can be flexibly set according to the application requirements of the actual scenario, and no specific limitation is made here.

[0073] Through the design of the above prompt words, it can be ensured that the generated collection summary information not only includes the core points in the collection process, but also meets the actual needs of the collection work, thereby improving the accuracy and efficiency of collection information generation and providing strong support for subsequent management and analysis.

[0074] In some embodiments of the present disclosure, generating reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information includes: constructing a second preset prompt word corresponding to the reminder result form information; obtaining a corresponding generation example of the reminder summary information and the reminder result form information; generating reminder result form information using the reminder information generation model according to the reminder summary information, the second preset prompt word corresponding to the reminder result form information, and the corresponding generation example of the reminder summary information and the reminder result form information.

[0075] When generating reminder result form information, the designed second preset prompt information may include, in addition to the prompt words corresponding to the reminder result form information, several examples of extracting the reminder result form information based on the reminder summary information may be designed and given, and the prompt words corresponding to the reminder result form information and the extracted examples may be input together as prompt information into the reminder information generation model. The reminder information generation model may then automatically extract the reminder result form information from the reminder summary information according to the prompt information, thereby improving the generation efficiency of the reminder information.

[0076] How to design the prompt words corresponding to the reminder result form information and the input and output examples for extracting the reminder result form information from the reminder summary can be flexibly configured by those skilled in the art according to the requirements, and no specific limitation is made here.

[0077] In some embodiments of the present disclosure, the reminder result form information is structured reminder result form information, and the structured reminder result form information includes at least one of a customer communication result label, a reason for overdue payment, an expected repayment time, and a customer attitude. Generating the reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information includes: filling in the reminder result form based on the structured reminder result form information.

[0078] The collection result form information generated by the above-mentioned embodiment is structured collection result form information, which may include, for example, customer communication result labels, overdue reasons, expected repayment time, and customer attitude, etc. By structuring the collection key information, the machine can automatically fill in the collection result form to form a complete, accurate and easy-to-understand collection record. Structured information is more intuitive and can be stored in a relational database to facilitate subsequent collection data analysis. Compared with manual filling, it greatly improves the work efficiency of collection personnel and effectively increases the number of customers reached by collection personnel every day.

[0079] In some embodiments of the present disclosure, the reminder information generation model is trained in the following manner: constructing reminder information generation sample data, wherein the training sample data is constructed based on historical collection call recording data and corresponding historical reminder information; obtaining a pre-trained large language model, and using the training sample data to fine-tune the pre-trained large language model to obtain a trained reminder information generation model.

[0080] Continue to refer Figure 2 In order to implement the aforementioned embodiments, it is necessary to train the reminder information generation model in advance. Considering that the amount of training sample data collected and labeled for directly training a large-scale reminder information generation model is very large, and the training takes a long time, the present disclosure adopts the method of pre-training + fine-tuning to train the required reminder information generation model.

[0081] Specifically, we first obtain historical call records and historical reminder data to build a training sample data set, and then use the training sample data set to fine-tune the general pre-trained large language model to improve the model effect, so as to further meet the requirements of reminder information generation. The pre-trained large language model can be implemented using the ChatGLM2-6B model, for example. ChatGLM2 has a wide range of application scenarios and can be deployed locally on consumer-grade graphics cards. For example, at the INT4 quantization level, only 6GB of video memory is required. In general, ChatGLM2-6B is a powerful, high-performance, and easy-to-deploy large language model that is suitable for real-time natural language processing tasks, especially for application scenarios that require high-performance language models but have limited hardware resources.

[0082] For large models, the number of parameters is very large, and retraining all model parameters to fine-tune becomes infeasible. There are high requirements for the scale of training data, machine resources and time consumption. If all parameters are not trained well, the original parameters of the large model will be disturbed, resulting in poor results. Therefore, after having a large language model as a base with sufficient general knowledge, when applying it to specific downstream tasks, a certain amount of labeled training data can be used to fine-tune the model instructions. Fine-tuning can allow the large model to better know how to call the "instilled" knowledge to complete specific field application tasks. The present disclosure is based on a self-constructed data set, and can be fine-tuned using the large model fine-tuning method LoRA based on extremely low resources. LoRA can use the data corresponding to the downstream task and adapt to the downstream task only by training the newly added parameters. After the new parameters are trained, the new parameters and the old model parameters are merged by re-parameterization, so as to achieve the effect of fine-tuning the entire model on the new task.

[0083] The main idea of ​​LoRA is to add a bypass next to the original pre-trained language model (PLM) to perform a dimensionality reduction and then dimensionality increase operation. During training, the parameters of the PLM are fixed, and only the dimensionality reduction matrix A and the dimensionality increase matrix B are trained, while the input and output dimensions of the model remain unchanged. When outputting, the matrices B, A and PLM parameters are superimposed.

[0084] Assuming that ChatGLM-6B needs to be fine-tuned, the parameters of the pre-trained large model need to be updated as W: W = W0 + ΔW, where W0 is the parameter initialized by the pre-trained model, and ΔW is the parameter that needs to be updated. If the ChatGLM-6B model is fine-tuned directly, the number of parameters to be trained ΔW is 6 billion, which is very expensive. For LoRA, during the training process, W0 is fixed, only A and B are training parameters, and only ΔW needs to be fine-tuned. Assuming that the pre-trained matrix is ​​W0∈Rd*k, its update can be expressed as W0+ΔW=W0+BA, B∈Rd*r, A∈Rr*k, where the rank r<<min(d,k), so the order of magnitude of fine-tuning parameters can be greatly reduced through low-rank decomposition.

[0085] In some embodiments of the present disclosure, the historical collection information includes historical collection summary information, and the construction of collection information generation sample data includes: obtaining historical collection call recording data and obtaining corresponding historical collection summary information; converting the historical collection call recording data into historical collection text data; and constructing collection information generation sample data based on the historical collection text data and the corresponding historical collection summary information.

[0086] The training of the collection information generation model in the disclosed embodiment is mainly to train the model's ability to generate collection summary information. Therefore, when constructing sample data for collection information generation, historical collection call recording data and corresponding historical collection summary information can be obtained first, and then the historical collection call recording data can be converted into historical collection text data using technologies such as speech recognition, and the historical collection text data and historical collection summary information can be used as sample data for collection information generation.

[0087] It should be noted that, on the one hand, the generation of reminder result form information is to generate reminder information in a table structure, and the difficulty and cost of additional training models to generate reminder information in a table structure are higher. On the other hand, considering the semantic similarity between reminder summary information and reminder result form information, reminder result form information can be further extracted based on reminder summary information. Therefore, the embodiment of the present disclosure only needs to fine-tune the model's ability to generate reminder summary information, and does not need to separately train the ability to generate reminder result form information, thereby reducing the model training cost and improving the training efficiency.

[0088] In some embodiments of the present disclosure, the historical reminder information also includes historical reminder result form information, and obtaining the corresponding historical reminder summary information includes: determining whether there is corresponding historical reminder summary information for the call recording data of the historical reminder; if so, directly obtaining the historical reminder summary information corresponding to the call recording data of the historical reminder; otherwise, generating the corresponding historical reminder summary information according to the historical reminder result form information.

[0089] In actual scenarios, after completing a collection task, general collection personnel will record the collection results in the collection system, which generally includes a collection summary and a structured collection result form. However, because the collection summary is usually a paragraph of text and is not a mandatory field, many collection personnel may not fill it out. The collection result form is generally a drop-down selection, and is generally required to be filled out by the collection personnel. Therefore, the collection form has richer and more complete information content.

[0090] Based on this, in order to meet the training sample data volume requirement during the historical collection stage, if the actual collection personnel do not fill in the collection summary data, the structured collection result form information can be spliced ​​together to supplement the collection summary data. In the above way, the training sample data set is constructed into the input and response text format samples required by the large language model, that is, the input is the call data after the recording is converted to text, and the response output is the corresponding collection summary data after data processing.

[0091] It should be noted here that most of the training data collected by the collection information generation model to generate collection summary information are summary texts formed by splicing historical collection result form information. Therefore, the summary text and the structured information to be extracted are very similar in semantics, and even the collection summary filled out by the collection personnel themselves will generally cover the content of the collection result form information. Therefore, this is also the case for the generation of collection result form information in the aforementioned embodiment. The prompt words can be directly designed and several extracted input and output examples can be given. Without any additional training or fine-tuning, a large model can be used for extraction through a small sample method (In-context learning). The accuracy is high, and there is no need to separately train the large language model for the generation of collection result form information, which reduces the cost of model training.

[0092] In summary, the reminder information generation method disclosed in the present invention has achieved at least the following technical effects:

[0093] 1) Apply the large-scale conversational language generation model to the collection scenario, and use the powerful natural language understanding, generation and key information extraction capabilities of the large model to automatically generate conversational collection summaries and collection result table information. Compared with the previous method of manual filling by collection personnel, it effectively improves the efficiency of collection personnel in recording collection results and increases the average number of collection cases handled per day.

[0094] 2) Compared with other models, large language models can reduce the reliance on large amounts of labeled data and use fine-tuning technology to improve the performance of general large models. The historical collection dialogue recordings and collection results are used as training data to fine-tune the model, which reduces the cost of data labeling and training calculations, mobilizes the capabilities of the general model for this downstream task, and significantly improves the performance of the model on the task.

[0095] 3) Generate a collection summary based on the call data during the collection process. On the one hand, it utilizes the text summary generation capability that the large model is good at; on the other hand, it splits the collection information generation task into two tasks: collection summary and structured collection result information extraction. The summary text containing key information is used to further extract structured collection result form data for auxiliary form filling. Compared with directly generating form data, it has better results and higher efficiency.

[0096] The present disclosure also provides a reminder information generating device 300, such as Figure 3 As shown, a schematic diagram of the structure of a reminder information generating device in an embodiment of the present disclosure is provided. The reminder information generating device 300 includes: an acquisition unit 310, a conversion unit 320, a first generation unit 330 and a second generation unit 340, wherein:

[0097] An acquisition unit 310 is used to acquire the collection call recording data;

[0098] A conversion unit 320, configured to convert the collection call recording data into collection text data;

[0099] A first generating unit 330 is used to generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model;

[0100] The second generating unit 340 is used to generate reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information.

[0101] In some embodiments of the present disclosure, the first generating unit 330 is specifically used to: construct a first preset prompt word corresponding to the reminder summary information; and generate the reminder summary information using the reminder information generation model according to the reminder text data and the first preset prompt word.

[0102] In some embodiments of the present disclosure, the second generating unit 340 is specifically used to: construct a second preset prompt word corresponding to the reminder result form information; obtain a corresponding generation example of the reminder summary information and the reminder result form information; and generate the reminder result form information using the reminder information generation model according to the reminder summary information, the second preset prompt word corresponding to the reminder result form information, and the corresponding generation example of the reminder summary information and the reminder result form information.

[0103] In some embodiments of the present disclosure, the reminder result form information is structured reminder result form information, and the structured reminder result form information includes at least one of a customer communication result label, a reason for overdue payment, an expected repayment time, and a customer attitude. The second generation unit 340 is specifically used to fill in the reminder result form based on the structured reminder result form information.

[0104] In some embodiments of the present disclosure, the reminder information generation model is trained in the following manner: constructing reminder information generation sample data, wherein the training sample data is constructed based on historical collection call recording data and corresponding historical reminder information; obtaining a pre-trained large language model, and using the training sample data to fine-tune the pre-trained large language model to obtain a trained reminder information generation model.

[0105] In some embodiments of the present disclosure, the historical collection information includes historical collection summary information, and the construction of collection information generation sample data includes: obtaining historical collection call recording data and obtaining corresponding historical collection summary information; converting the historical collection call recording data into historical collection text data; and constructing collection information generation sample data based on the historical collection text data and the corresponding historical collection summary information.

[0106] In some embodiments of the present disclosure, the historical reminder information also includes historical reminder result form information, and obtaining the corresponding historical reminder summary information includes: determining whether there is corresponding historical reminder summary information for the call recording data of the historical reminder; if so, directly obtaining the historical reminder summary information corresponding to the call recording data of the historical reminder; otherwise, generating the corresponding historical reminder summary information according to the historical reminder result form information.

[0107] It can be understood that the reminder information generating device described above can implement each step of the reminder information generating method provided in the above-mentioned embodiment. The relevant explanations on the reminder information generating method are applicable to the reminder information generating device and will not be repeated here.

[0108] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 4 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0109] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0110] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0111] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a reminder information generation device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0112] Obtain collection call recording data;

[0113] Converting the collection call recording data into collection text data;

[0114] Generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model;

[0115] According to the reminder summary information and the second preset prompt information, the reminder information generation model is used to generate reminder result form information.

[0116] The above application Figure 1 The method performed by the reminder information generation device disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0117] The present disclosure also provides a computer program product, including a computer program or an instruction, which is implemented when the computer program or the instruction is executed by a processor. Figure 1 The reminder information generation method in the embodiment shown is specifically used to execute:

[0118] Obtain collection call recording data;

[0119] Converting the collection call recording data into collection text data;

[0120] Generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model;

[0121] According to the reminder summary information and the second preset prompt information, the reminder information generation model is used to generate reminder result form information.

[0122] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0125] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0126] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0127] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0128] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0129] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0130] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0131] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for generating reminder information, characterized in that: The reminder information generation method comprises: Obtain collection call recording data; Converting the collection call recording data into collection text data; Generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model; According to the reminder summary information and the second preset prompt information, the reminder information generation model is used to generate reminder result form information.

2. The method for generating reminder information according to claim 1, characterized in that: The generating of reminder summary information by using a reminder information generation model according to the reminder text data and the first preset prompt information includes: Constructing a first preset prompt word corresponding to the reminder summary information; The reminder summary information is generated by using the reminder information generation model according to the reminder text data and the first preset prompt word.

3. The method for generating reminder information according to claim 1, characterized in that: The generating of reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information includes: Constructing a second preset prompt word corresponding to the reminder result form information; Obtain the corresponding generation example of reminder summary information and reminder result form information; The reminder result form information is generated using the reminder information generation model according to the reminder summary information, the second preset prompt word corresponding to the reminder result form information, and the corresponding generation example of the reminder summary information and the reminder result form information.

4. The method for generating reminder information according to claim 1, characterized in that: The reminder result form information is structured reminder result form information, and the structured reminder result form information includes at least one of a customer communication result label, a reason for overdue payment, an expected repayment time, and a customer attitude. The reminder result form information is generated by using the reminder information generation model according to the reminder summary information and the second preset prompt information, including: Fill in a reminder result form based on the structured reminder result form information.

5. The method for generating reminder information according to any one of claims 1 to 4, characterized in that: The reminder information generation model is trained in the following way: Constructing collection reminder information to generate sample data, wherein the training sample data is constructed based on historical collection call recording data and corresponding historical collection reminder information; A pre-trained large language model is obtained, and the pre-trained large language model is fine-tuned using the training sample data to obtain a trained reminder information generation model.

6. The method for generating reminder information according to claim 5, characterized in that: The historical reminder information includes historical reminder summary information, and the construction of reminder information to generate sample data includes: Obtain historical collection call recording data and corresponding historical collection summary information; Converting the historical debt collection call recording data into historical debt collection text data; The reminder information generation sample data is constructed according to the historical reminder text data and the corresponding historical reminder summary information.

7. The method for generating reminder information according to claim 6, characterized in that: The historical reminder information also includes historical reminder result form information, and obtaining the corresponding historical reminder summary information includes: Determine whether the historical collection call recording data has corresponding historical collection summary information; If it exists, directly obtain the historical collection summary information corresponding to the call recording data of the historical collection; Otherwise, the corresponding historical reminder summary information is generated according to the historical reminder result form information.

8. A reminder information generating device, characterized in that: The reminder information generating device comprises: An acquisition unit, used to acquire collection call recording data; A conversion unit, used to convert the collection call recording data into collection text data; A first generating unit, configured to generate reminder summary information using a reminder information generation model according to the reminder text data and the first preset prompt information, wherein the reminder information generation model is obtained by fine-tuning a pre-trained large language model; The second generating unit is used to generate reminder result form information using the reminder information generation model according to the reminder summary information and the second preset prompt information.

9. An electronic device, comprising: processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes the reminder information generating method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the method for generating reminder information according to any one of claims 1 to 7 is implemented.

Images