Method and device for interacting with character customer service, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of artificial intelligence, and discloses a method and device for interacting with character customer service, electronic equipment and a storage medium. The method comprises the following steps: acquiring user input data, and preprocessing the user input data to obtain target input data; inputting the target input data into a target language representation model generated by pre-training to obtain prompt and supplementary suggestion information input to the user; performing fuzzy recognition on the user input data to obtain fuzzy input data; according to prompt and supplementary suggestion information corresponding to the fuzzy input data, uncertainty measurement of model identification is carried out; and when the uncertainty measurement result meets a model uncertainty optimization condition, performing model optimization on the target language representation model, performing character customer service interaction by adopting the optimized target language representation model, and performing character customer service interaction by optimizing the model when the fuzzy input data cannot be accurately recognized by the model. The fuzzy input data processing performance of the model can be improved, and the recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, electronic device and storage medium for interacting with text customer service. Background Art

[0002] With the development of artificial intelligence technology, when dealing with user problems, text customer service is usually used to connect with users and resolve the issues.

[0003] In the prior art, the user input is often matched with the templates in the knowledge base to find the most similar template and then return the corresponding preset response. The prior art method is effective for dealing with common problems, but for new or complex problems, the system is limited in understanding and responding to user input due to the lack of fuzzy context and context information, resulting in the prompts or responses generated by the system being inaccurate or not relevant.

[0004] Therefore, there is an urgent need to provide a method for interacting with text customer service to improve the ability to process fuzzy contexts in order to improve the accuracy of user intent recognition. Summary of the invention

[0005] The present invention provides a method, device, electronic device and storage medium for interacting with text customer service to improve fuzzy context processing capability and user intention recognition accuracy.

[0006] According to one aspect of the present invention, a method for interacting with text customer service is provided, the method comprising:

[0007] Acquire user input data, and preprocess the user input data to obtain target input data;

[0008] Inputting the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input;

[0009] Performing fuzzy recognition on the user input data to obtain fuzzy input data; and performing uncertainty measurement of model recognition according to the prompt and supplementary suggestion information corresponding to the fuzzy input data;

[0010] When the uncertainty measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, and the optimized target language representation model is used for text customer service interaction.

[0011] According to another aspect of the present invention, a device for interacting with text customer service is provided, the device comprising:

[0012] A target input data determination module is used to obtain user input data and pre-process the user input data to obtain target input data;

[0013] A prompt and supplementary suggestion information determination module, used to input the target input data into the target language representation model generated by pre-training to obtain prompt and supplementary suggestion information for the user input;

[0014] An uncertainty measurement module, used to perform fuzzy recognition on the user input data to obtain fuzzy input data; and perform uncertainty measurement of model recognition according to the prompts and supplementary suggestion information corresponding to the fuzzy input data;

[0015] The model optimization module is used to optimize the target language representation model when the uncertainty measurement result meets the model uncertainty optimization condition, and use the optimized target language representation model for text customer service interaction.

[0016] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for interacting with text customer service described in any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method of interacting with text customer service described in any embodiment of the present invention when executed.

[0021] According to another aspect of the present invention, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the method for interacting with text customer service described in any embodiment of the present invention is implemented.

[0022] The technical solution of the embodiment of the present invention obtains user input data, pre-processes the user input data, and obtains target input data; inputs the target input data into a target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input; fuzzy recognizes the user input data to obtain fuzzy input data; and measures the uncertainty of model recognition based on the prompts and supplementary suggestion information corresponding to the fuzzy input data; when the uncertainty measurement result meets the model uncertainty optimization condition, optimizes the target language representation model, and uses the optimized target language representation model for text customer service interaction, which solves the problem of insufficient fuzzy context processing ability when interacting with text customer service. By optimizing the model when the model cannot accurately recognize the fuzzy input data, the model's processing performance on the fuzzy input data can be improved, and the recognition accuracy of the model can be improved.

[0023] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 is a flowchart of a method for interacting with text customer service provided according to Embodiment 1 of the present invention;

[0026] Figure 2 is a flowchart of a method for interacting with text customer service provided according to Embodiment 2 of the present invention;

[0027] Figure 3 is a schematic diagram of the structure of a device for interacting with text customer service provided according to Embodiment 3 of the present invention;

[0028] Figure 4 It is a structural schematic diagram of an electronic device for implementing the method for interacting with text customer service according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Embodiment 1

[0032] Figure 1 This is a flowchart of a method for interacting with text customer service according to the first embodiment of the present invention. This embodiment is applicable to the situation where a user interacts with text customer service in a banking business. The method can be executed by a device for interacting with text customer service. The device for interacting with text customer service can be implemented in the form of hardware and / or software. The device for interacting with text customer service can be configured in an electronic device. The electronic device can be a mobile device such as a mobile phone, a tablet computer (PAD), a wearable device, or a personal computer (PC, Personal Computer), etc. Figure 1 As shown, the method includes:

[0033] Step 110: Obtain user input data, and pre-process the user input data to obtain target input data.

[0034] The acquisition of personal information is only done after authorization by the user, and the acquisition method is reasonable and legal. User input data may include questions that the user asks the text customer service in advance. For example, user input data may include questions that the user asks the text customer service of the bank. User input data may also include contextual information related to the question, background information and other data.

[0035] The preprocessing of user input data can include word segmentation, word embedding, and context integration. Preprocessing user input data can enable the target language representation model to better understand user questions.

[0036] Step 120: Input the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input.

[0037] The target language representation model can be generated by training based on the basic language representation model and the relevant data of interaction with text customer service. Specifically, the target language representation model can be generated by training based on the basic language representation model and the relevant data of interaction with text customer service of the bank. The basic language representation model can be a Transformer-based bidirectional encoder representation (Bidirectional Encoder Representations from Transformers, BERT) model in natural language processing (NLP). The BERT model contains multiple encoder layers, each of which can transform and encode the input to generate a higher-level semantic representation. This multi-layer representation learning helps to capture complex contextual information, has a better understanding of user input, and has the ability of context perception. The BERT model can take into account historical conversations, user preferences, and the current conversation context to more accurately understand user needs and generate more targeted prompts and additional suggestions.

[0038] The BERT model can achieve semantic understanding of user input data. For example, user input data can be segmented, embedded, and combined with context to ensure that the user's intentions and needs are accurately grasped.

[0039] In order to improve the timeliness of model recognition response, a lightweight model loading method can be used. For example, in order to enable the target language representation model to complete the reasoning process in a short time, an efficient reasoning algorithm can be used, through GPU hardware acceleration and quantizing the weight of the target language representation model from 32-bit floating point to 8-bit integer, which significantly reduces memory usage and calculation time, reduces calculation complexity, and improves the real-time performance of the system.

[0040] The target language representation model can understand the target input data, comprehensively consider the context, and obtain corresponding prompts and supplementary suggestion information. The prompts and supplementary suggestion information can be returned to the user in text form, realizing instant automatic interaction.

[0041] Among them, based on the semantic understanding of the BERT model, relevant prompts are generated, which can provide information or suggestions of interest to users. The model encodes the user input and then decodes it to generate relevant prompts, which can cover the topic of the user input and take into account the historical conversation in context to provide more personalized and comprehensive suggestions. Suggestions can involve details, related questions, or further information. The generation of supplementary suggestions is based on the model's in-depth understanding of the user input data and context, which can ensure the provision of comprehensive and targeted information.

[0042] Step 130: Perform fuzzy recognition on the user input data to obtain fuzzy input data; and perform uncertainty measurement of model recognition based on the prompts and supplementary suggestion information corresponding to the fuzzy input data.

[0043] Since users inevitably have ambiguous input when interacting with text customer service, in order to improve the robustness of interaction with text customer service, an error handling mechanism is introduced in the embodiment of the present invention. The error handling mechanism performs fuzzy recognition on the user input data to determine the fuzzy input data in the user input data. The fuzzy input data includes but is not limited to user input data that is ambiguous, uncertain, or cannot be determined by the model.

[0044] In an embodiment of the present invention, rules may be defined for fuzzy recognition. Specifically, fuzzy recognition may include but is not limited to recognizing user input that may cause ambiguity or uncertainty. Exemplarily, fuzzy recognition may include abbreviation errors, spelling errors, and ambiguous expressions.

[0045] For inputs identified as fuzzy or uncertain, the embodiment of the present invention introduces uncertainty measurement. The uncertainty measurement can be determined by measuring one-dimensional or multi-dimensional indicators such as the degree of understanding of the target language representation model to the user input, the response speed, and the user's satisfaction with the response result. For example, when the prompt and supplementary suggestion information corresponding to the fuzzy input data has a low confidence level, it can be indicated that the model has uncertainty when processing the fuzzy input data.

[0046] In an optional implementation of an embodiment of the present invention, uncertainty measurement of model identification is performed based on prompts and supplementary suggestion information corresponding to the fuzzy input data, including: uncertainty measurement of model identification based on probability distribution, confidence, user feedback data, model inference time, processor usage, and memory usage in the prompts and supplementary suggestion information corresponding to the fuzzy input data.

[0047] Among them, the probability distribution and confidence level in the prompt and supplementary suggestion information corresponding to the fuzzy input data can be information related to the credibility of the output result outputted synchronously by the target language representation model when outputting the specific content of the prompt and supplementary suggestion information. The user feedback data can be the data provided by the user based on the prompt and supplementary suggestion information when the target language representation model outputs the prompt and supplementary suggestion information. For example, the user feedback data can be further supplementary information inputted by the user for the prompt and supplementary suggestion information, or jump instructions such as contacting manual customer service, etc.

[0048] The model reasoning time may be the time required for the model to output prompts and supplementary suggestion information obtained by running the target language representation model. For example, the model reasoning time may be measured by the Python time module to evaluate the real-time performance of the model reasoning. The processor usage and memory usage may be the processor and memory usage when the model processes the target input data, outputs prompts and supplementary suggestion information, obtained by running the target language representation model. For example, the processor (such as CPU) and memory usage may be monitored by a performance monitoring tool (such as the psutil library).

[0049] The uncertainty measurement of model recognition can be specifically to integrate the probability distribution, confidence, user feedback data, model reasoning time, processor usage, and memory usage, and to evaluate the reliability of the prompts and supplementary suggestion information obtained by model recognition. The lower the probability distribution and confidence, the greater the uncertainty. The more user feedback data or the higher the frequency of requesting manual customer service, the greater the uncertainty. The longer the model reasoning time, the greater the uncertainty. The higher the processor usage and memory usage ratio, the greater the uncertainty.

[0050] For example, preset weights can be set for the parameters of each dimension, and the probability distribution, confidence, user feedback data, model reasoning time, processor usage, and memory usage are mapped to uncertainty, and weighted summation is performed according to the corresponding weights to obtain the uncertainty measurement value of model recognition. Among them, the weight values ​​of probability distribution, confidence, and user feedback data can be set to larger values; model reasoning time, processor usage, and memory usage can be set to smaller weight values.

[0051] By performing multi-dimensional uncertainty measurement on model recognition, the performance of fuzzy input processing in model recognition can be evaluated, and then it can be determined whether the target language representation model needs to be optimized. The timing of model optimization can be accurately determined, and the timeliness of model optimization can be improved, so that the model can accurately process fuzzy input, thereby improving the accuracy of model recognition and maintaining the recognition efficiency and effectiveness of the model in practical applications.

[0052] Step 140: When the uncertainty measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, and the optimized target language representation model is used for text customer service interaction.

[0053] The uncertainty measurement result meeting the model uncertainty optimization condition can mean that the uncertainty measurement triggers the model uncertainty optimization. For example, when the probability distribution and confidence level are lower than the preset credibility value, the model can be uncertainty optimized. The more user feedback data or the higher the frequency of requesting manual customer service, the more uncertainty optimization can be performed on the model. The longer the model reasoning time, the more uncertainty optimization can be performed on the model. The higher the processor usage and memory usage ratio, the more uncertainty optimization can be performed on the model. In specific applications, one or more of the above situations can be combined to improve the accuracy of the timing of uncertainty optimization of the model. For example, the uncertainty measurement value obtained by weighted summing of the parameters mentioned above can be used for judgment. When the uncertainty measurement value is greater than the preset uncertainty threshold, the model can be uncertainty optimized.

[0054] Uncertainty optimization of the model can include a variety of methods. For example, uncertainty optimization of the model can be incremental learning of the target language representation model, and / or, optimization of the target language representation model such as improving reasoning efficiency, to ensure that the target language representation model can respond accurately and quickly. After optimizing the target language representation model, the optimized target language representation model can be used in subsequent interactions with text customer service to improve the accuracy and speed of model recognition.

[0055] In an optional implementation of an embodiment of the present invention, when the uncertain measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, including: when the uncertain measurement result meets the model uncertainty optimization condition, obtaining user feedback and context information associated with the fuzzy input data, and performing incremental learning on the target language representation model; optimizing the operating efficiency of the target language representation model through an inference engine.

[0056] Exemplarily, incremental learning of a target language representation model may include the following process: loading a pre-trained target language representation model; obtaining user feedback and context information associated with fuzzy input data; performing incremental learning through a Trainer tool (a tool in machine learning) to build a Trainer instance; and starting incremental learning training.

[0057] By performing incremental learning on the target language representation model, the model can be dynamically adjusted so that it can continuously learn new data, adapt to user feedback, context changes, and new information learned, thereby improving the model's adaptability. This enables the generation strategy of prompts and supplementary suggestions to be adjusted in real time, maintaining the practicality and timeliness of model recognition.

[0058] On the basis of incremental learning, the inference engine of the target language representation model can also be optimized to improve the operating efficiency of the model. For example, the operating efficiency of the model can be optimized through the TensorRT (a neural network optimization toolkit) inference engine to ensure fast and accurate responses when users interact with text customer service.

[0059] The technical solution of this embodiment obtains user input data, pre-processes the user input data, and obtains target input data; inputs the target input data into a target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input; fuzzy recognizes the user input data to obtain fuzzy input data; and measures the uncertainty of model recognition based on the prompts and supplementary suggestion information corresponding to the fuzzy input data; when the uncertainty measurement result meets the model uncertainty optimization condition, optimizes the target language representation model, and uses the optimized target language representation model for text customer service interaction, which solves the problem of insufficient fuzzy context processing ability when interacting with text customer service. By optimizing the model when the model cannot accurately recognize the fuzzy input data, the model's processing performance on the fuzzy input data can be improved, and the recognition accuracy of the model can be improved.

[0060] Embodiment 2

[0061] Figure 2 1 is a flowchart of a method for interacting with text customer service according to Embodiment 2 of the present invention. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with each optional solution in one or more of the above embodiments. Figure 2 As shown, the method includes:

[0062] Step 210: Obtain user input data, and pre-process the user input data to obtain target input data.

[0063] Step 220: Input the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input.

[0064] In an optional implementation of an embodiment of the present invention, before inputting the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input, it also includes: obtaining a labeled intent classification dataset, and iteratively training the basic language representation model to obtain the target language representation model; in the iterative training, the parameters of the basic language representation model are adjusted through the similarity between the model output results and the intent in the intent classification dataset, and the generalization ability of the model on unseen data.

[0065] The labeled intent classification dataset can be data related to banking services, generated by labeling. The labeled intent classification dataset can include diverse user inputs, text customer service responses, and contextual information. By including typical user query information in banking services in the labeled intent classification dataset, the model can learn a wide range of contexts and use cases during fine-tuning training.

[0066] In order to make the target language representation model better adapt to the text customer service interaction task, a loss function can be constructed during training to adjust parameters. Among them, the loss function can reflect the similarity between the model output results and the intent in the intent classification dataset, as well as the generalization ability of the model on unseen data. Specifically, the loss function can adopt the cross entropy loss function. The similarity between the model output results and the intent in the intent classification dataset can be determined by considering the difference between the probability distribution predicted by the model and the probability distribution of the true label. For example, the loss function can be to maximize the vocabulary information in the context and minimize the difference between the user input and the expected output, so that the model can better understand the user's questions and generate relevant prompts and additional suggestions.

[0067] For example, for a given input x, assume that the probability distribution predicted by the model is P model (y|x). The probability distribution of the true label is P ture (y|x), then the cross entropy loss function is: In practical applications, when the task is a classification problem, y is a specific category label, and the cross entropy loss can be simplified as: Among them, y i is the true label (usually one-hot), is the probability predicted by the model.

[0068] The generalization ability of the model on unseen data can be verified through the test set, and the generalization ability of the model can be determined based on the accuracy of the model prediction to adjust the model parameters.

[0069] Based on the data set and loss function, fine-tuning the model can make the model better adapted to the specific field of text customer service. The fine-tuning process can use the Stochastic Gradient Descent (SGD) optimization algorithm to continuously adjust the model parameters through multiple iterations on the data set to make it better adapted to the specific field of text customer service.

[0070] The embodiment of the present invention can learn a common representation from a large amount of training data by adjusting and training the basic language representation model, and has better generalization ability for unseen data. The generated target language representation model can understand the deep semantics of the sentence, not just the surface similarity; it has the ability to handle complex semantics, and can understand the ambiguity of words through context, thereby improving the accuracy and naturalness of automatic prompts and supplements, and can be applied to text customer service interactions in specific scenarios, so that it has stronger semantic understanding and user intent recognition capabilities, thereby improving user experience. And the performance of the target language representation model can be further improved by adding training data and adjusting the model structure (such as increasing the number of layers, changing the learning rate, etc.), and rich open source libraries and tools can be used for rapid iteration and optimization.

[0071] The trained target language representation model can be saved and deployed. In actual applications, real-time reasoning performance can be achieved by inputting user input data into the target language representation model, and instant automatic prompts and supplementary suggestions for interacting with text customer service can be obtained.

[0072] After generating prompts and supplementary suggestion information, the user may be prompted to input feedback data. The prompt and supplementary suggestion generation algorithm may be further optimized through the user feedback data to adapt to the user's personalized needs, thus forming an iterative learning process.

[0073] Step 230: Perform fuzzy recognition on the user input data to obtain fuzzy input data; and perform uncertainty measurement of model recognition based on the prompts and supplementary suggestion information corresponding to the fuzzy input data.

[0074] In an optional implementation of an embodiment of the present invention, uncertainty measurement of model identification is performed based on prompts and supplementary suggestion information corresponding to the fuzzy input data, including: uncertainty measurement of model identification based on probability distribution, confidence, user feedback data, model inference time, processor usage, and memory usage in the prompts and supplementary suggestion information corresponding to the fuzzy input data.

[0075] Step 240: Generate a prompt redirection word according to the uncertainty measurement result, the user input data, the background information associated with the user input data, and the intention data.

[0076] When it is determined that there is uncertainty in the user input data, the electronic device may prompt the user with a redirection word. The prompt redirection word may guide the user to provide more background information, detailed description or clarify the intention. For example, you may prompt: "Please provide more context so that I can better understand your question." In order to enable the user to know more clearly what information to input when prompting the user, the embodiment of the present invention may combine the user input data, the background information associated with the user input data, and the intention data when generating the prompt redirection word. For example, you may prompt: "Your input is identified as XX business. Please provide the YY data required in the XX business so that I can better understand your problem." For another example, when prompting a redirection word, relevant documents may be provided to guide the user to input a clearer expression.

[0077] Step 250: prompt the user to update the input data according to the prompt redirection word, and use the target language representation model to recognize the updated input data to obtain prompts and supplementary suggestion information.

[0078] Among them, after prompting the redirection word, the updated input data, as well as the corresponding prompts and supplementary suggestion information, can all be recorded to optimize the model. In the current interaction, the user can update the input data by prompting the redirection word, and then obtain the prompt and supplementary suggestion information of the current interaction. On the user application side, the current interaction can be ended after executing step 250. However, in order to ensure that the subsequent model has fuzzy processing capabilities, the model can be optimized, that is, step 260 is executed. Step 260 can be executed after executing step 230 or after step 250.

[0079] Step 260: When the uncertainty measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, and the optimized target language representation model is used for text customer service interaction.

[0080] In an optional implementation of an embodiment of the present invention, when the uncertain measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, including: when the uncertain measurement result meets the model uncertainty optimization condition, obtaining user feedback and context information associated with the fuzzy input data, and performing incremental learning on the target language representation model; optimizing the operating efficiency of the target language representation model through an inference engine.

[0081] The error handling mechanism is not just a one-time process. Through continuous learning and optimization, the model's ability to handle fuzzy or uncertain inputs can be continuously improved. By collecting user feedback, adjusting model parameters, and updating fuzzy input rules, the level of handling uncertainty can be gradually improved, thereby improving the overall recognition performance.

[0082] Step 270: Generate multimodal prompt information for display based on the prompt and supplementary suggestion information output by the optimized target language representation model and the type of user input data.

[0083] Among them, the multimodal prompt information may include but is not limited to content in rich media forms such as text, voice, image, and link. When generating the multimodal prompt information, the consistency with the type of data input by the user can be maintained. For example, when the user input data is text, multimodal prompt information such as text, image, and link can be generated. For another example, when the user input data is voice, multimodal prompt information such as voice, video, image, text, and link can be generated. Through multimodal information display, it is possible to interact with users in a richer and more intuitive way, thereby improving the user experience.

[0084] Step 280: Cache the prompts and supplementary suggestion information obtained by text customer service interaction using the optimized target language representation model, as well as the corresponding user input data, into a cache.

[0085] In an embodiment of the present invention, the next time the user interacts with text customer service, the target prompt and supplementary suggestion information corresponding to the next user input data is queried in the cache; when the target prompt and supplementary suggestion information are queried, the target prompt and supplementary suggestion information are input; otherwise, the optimized target language representation model is continued to be used for text customer service interaction.

[0086] Through caching technology, common user queries can be responded to quickly directly through the cache, avoiding repeated calculations using the target language representation model. This can save resources on the one hand, and improve the response rate when interacting with text customer service on the other.

[0087] The technical solution of the embodiment of the present invention obtains user input data and pre-processes the user input data to obtain target input data; inputs the target input data into a pre-trained target language representation model to obtain prompts and supplementary suggestion information for the user input; performs fuzzy recognition on the user input data to obtain fuzzy input data; and measures the uncertainty of model recognition based on the prompts and supplementary suggestion information corresponding to the fuzzy input data; generates prompt redirection words based on the user input data, background information associated with the user input data, and intention data; prompts the user to update the input data based on the prompt redirection words, and uses the target language representation model to The updated input data is identified to obtain prompts and supplementary suggestion information; when the uncertainty measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, and the optimized target language representation model is used for text customer service interaction; according to the prompts and supplementary suggestion information output by the optimized target language representation model, as well as the type of user input data, multimodal prompt information is generated for display; the prompts and supplementary suggestion information obtained by using the optimized target language representation model for text customer service interaction, and the corresponding user input data are cached in the cache, and the next time the text customer service is interacted with, the cache is queried for the prompts and supplementary suggestion information corresponding to the next user. The target prompt and supplementary suggestion information corresponding to the user input data are obtained; the next time the text customer service is interacted with, the target prompt and supplementary suggestion information corresponding to the next user input data are queried in the cache, which solves the problem of insufficient fuzzy context processing ability when interacting with text customer service. By fine-tuning the BERT model, context information can be captured more comprehensively, and a more accurate understanding of user input can be achieved, thereby improving the accuracy of identifying user intentions and making prompts more personalized; through real-time reasoning, more immediate automatic prompts and supplementary suggestions can be obtained when interacting with text customer service, which improves the real-time nature of the interaction; by dynamically adjusting the generation strategy, according to the user's historical By integrating conversations, preferences and real-time feedback, it provides prompts that better meet user needs, thereby improving the personalization level of user experience; by introducing a detailed error handling mechanism, it can guide users to provide more specific information in a friendly manner when faced with ambiguous or uncertain user input, thereby improving the efficiency, accuracy, and robustness of handling uncertainty; through continuous learning and optimization mechanisms, it can continuously improve the ability of automatic prompts and supplements, has stronger adaptability, and can maintain efficient performance in the ever-changing user needs and contexts; through multimodal prompts, the prompt generation is more rich, flexible, intuitive and vivid, enhancing the interactive experience between users and text customer service.

[0088] Embodiment 3

[0089] Figure 3 FIG. 1 is a schematic diagram of a structure of a device for interacting with text customer service according to Embodiment 3 of the present invention. Figure 3As shown, the device includes: a target input data determination module 310, a prompt and supplementary suggestion information determination module 320, an uncertainty measurement module 330, and a model optimization module 340.

[0090] in:

[0091] The target input data determination module 310 is used to obtain user input data and pre-process the user input data to obtain target input data;

[0092] A prompt and supplementary suggestion information determination module 320 is used to input the target input data into the target language representation model generated by pre-training to obtain prompt and supplementary suggestion information for the user input;

[0093] The uncertainty measurement module 330 is used to perform fuzzy recognition on the user input data to obtain fuzzy input data; and to perform uncertainty measurement of model recognition according to the prompts and supplementary suggestion information corresponding to the fuzzy input data;

[0094] The model optimization module 340 is used to optimize the target language representation model when the uncertainty measurement result meets the model uncertainty optimization condition, and use the optimized target language representation model for text customer service interaction.

[0095] Optionally, the uncertainty measurement module 330 includes:

[0096] The uncertainty measurement unit is used to measure the uncertainty of model recognition according to the probability distribution, confidence, user feedback data, model inference time, processor usage and memory usage in the prompt and supplementary suggestion information corresponding to the fuzzy input data.

[0097] Optionally, the model optimization module 340 includes:

[0098] An incremental learning unit, used to obtain user feedback and context information associated with the fuzzy input data and perform incremental learning on the target language representation model when the uncertain measurement result satisfies the model uncertainty optimization condition;

[0099] The operation efficiency improvement unit is used to optimize the operation efficiency of the target language representation model through the inference engine.

[0100] Optionally, the device further includes:

[0101] A prompt redirection word generation module, for generating prompt redirection words according to the uncertainty measurement result, user input data, background information associated with the user input data, and intention data after performing uncertainty measurement of model recognition according to the prompt and supplementary suggestion information corresponding to the fuzzy input data;

[0102] The re-recognition module is used to prompt the user to update the input data according to the prompt redirection word, and use the target language representation model to recognize the updated input data to obtain prompts and supplementary suggestion information.

[0103] Optionally, the device further includes:

[0104] A data cache module is used to cache the prompts and supplementary suggestion information obtained from text customer service interaction using the optimized target language representation model, as well as the corresponding user input data into a cache;

[0105] A cache query module, used to query the cache for target prompts and supplementary suggestion information corresponding to the next user input data when interacting with the text customer service next time;

[0106] The target prompt and supplementary suggestion information generation module is used to input the target prompt and supplementary suggestion information when the target prompt and supplementary suggestion information are queried; otherwise, the optimized target language representation model is continued to be used for text customer service interaction.

[0107] Optionally, the device further includes:

[0108] The multimodal prompt information generation module is used to generate multimodal prompt information for display according to the prompts and supplementary suggestion information output by the optimized target language representation model and the type of user input data after using the optimized target language representation model for text customer service interaction.

[0109] Optionally, the device further includes:

[0110] The target language representation model training module is used to obtain a labeled intent classification data set and iteratively train the basic language representation model to obtain the target language representation model before inputting the target input data into the pre-trained target language representation model to obtain prompts and supplementary suggestion information for the user input;

[0111] The parameter adjustment module is used to adjust the parameters of the basic language representation model during iterative training based on the similarity between the model output results and the intent in the intent classification dataset, as well as the generalization ability of the model on unseen data.

[0112] The device for interacting with text customer service provided in the embodiment of the present invention can execute the method for interacting with text customer service provided in any embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method.

[0113] In the technical solution of the embodiment of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of user personal information (such as user input data, user feedback data, and data in the interaction between users and text customer service, etc.) are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0114] The information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0115] When providing text-based customer service interaction services through the target language representation model, the corresponding operation entry is provided for the user to choose to agree or reject the automated decision results; if the user chooses to reject, the expert decision process is entered.

[0116] Embodiment 4

[0117] Figure 4 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0118] like Figure 4 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0119] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0120] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for interacting with text customer service.

[0121] In some embodiments, the method for interacting with text customer service may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for interacting with text customer service described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the method for interacting with text customer service in any other appropriate manner (e.g., by means of firmware).

[0122] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0123] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0124] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0125] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0126] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0127] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0128] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0129] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for interacting with text customer service, characterized in that: include: Acquire user input data, and preprocess the user input data to obtain target input data; Inputting the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input; Performing fuzzy recognition on the user input data to obtain fuzzy input data; and performing uncertainty measurement of model recognition according to the prompt and supplementary suggestion information corresponding to the fuzzy input data; When the uncertainty measurement result meets the model uncertainty optimization condition, the target language representation model is optimized, and the optimized target language representation model is used for text customer service interaction.

2. The method according to claim 1, characterized in that: According to the prompts and supplementary suggestion information corresponding to the fuzzy input data, uncertainty measurement of model identification is performed, including: Uncertainty measurement of model recognition is performed based on probability distribution, confidence, user feedback data, model inference time, processor usage, and memory usage in the prompts and supplementary suggestion information corresponding to the fuzzy input data.

3. The method according to claim 1, characterized in that When the uncertainty measurement result satisfies the model uncertainty optimization condition, the target language representation model is optimized, including: When the uncertainty measurement result satisfies the model uncertainty optimization condition, obtaining user feedback and context information associated with the fuzzy input data, and performing incremental learning on the target language representation model; The operating efficiency of the target language representation model is optimized through an inference engine.

4. The method according to claim 1, characterized in that: After uncertainty measurement of model identification is performed according to the prompts and supplementary suggestion information corresponding to the fuzzy input data, the method further includes: generating a prompt redirection word according to the uncertainty measurement result, the user input data, background information associated with the user input data, and the intention data; According to the prompt redirection word, the user is prompted to update the input data, and the updated input data is recognized using the target language representation model to obtain prompts and supplementary suggestion information.

5. The method according to claim 1, characterized in that Also includes: Cache the prompts and supplementary suggestion information obtained from text customer service interaction using the optimized target language representation model, as well as the corresponding user input data, in a cache; When interacting with the text customer service next time, querying the cache for target prompts and supplementary suggestion information corresponding to the next user input data; When the target prompt and the supplementary suggestion information are found, the target prompt and the supplementary suggestion information are input; otherwise, the optimized target language representation model is continuously used for text customer service interaction.

6. The method according to claim 1, characterized in that After using the optimized target language representation model for text customer service interaction, it also includes: Based on the prompts and supplementary suggestion information output by the optimized target language representation model and the type of user input data, multimodal prompt information is generated for display.

7. The method according to claim 1, characterized in that Before inputting the target input data into the target language representation model generated by pre-training to obtain prompts and supplementary suggestion information for the user input, the method further includes: Obtain a labeled intent classification dataset and iteratively train the basic language representation model to obtain the target language representation model; During iterative training, the parameters of the basic language representation model are adjusted based on the similarity between the model output results and the intent in the intent classification dataset, as well as the model's generalization ability on unseen data.

8. A device for interacting with text customer service, characterized in that: include: A target input data determination module is used to obtain user input data and pre-process the user input data to obtain target input data; A prompt and supplementary suggestion information determination module, used to input the target input data into the target language representation model generated by pre-training to obtain prompt and supplementary suggestion information for the user input; An uncertainty measurement module, used for performing fuzzy recognition on the user input data to obtain fuzzy input data; and performing uncertainty measurement of model identification according to the prompts and supplementary suggestion information corresponding to the fuzzy input data; The model optimization module is used to optimize the target language representation model when the uncertainty measurement result meets the model uncertainty optimization condition, and use the optimized target language representation model for text customer service interaction.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for interacting with text customer service according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for interacting with text customer service according to any one of claims 1 to 7 when executed.

Images