Mail reply method and device and electronic equipment

Abstract

The invention provides a mail reply method and device and electronic equipment. The method comprises the following steps: determining a mail reply template corresponding to a mail type and an emotion analysis result of a to-be-processed mail text; generating mail reply content based on the mail reply template and the intention recognition result of the to-be-processed mail text; and replying the mail based on the mail reply content. According to the method and the device provided by the invention, the mail reply template corresponding to the mail type and the sentiment analysis result of the to-be-processed mail text is determined; the mail reply content is generated based on the mail reply template and the intention recognition result of the to-be-processed mail text, so that the accuracy and specialty of automatic reply of the mail are improved, the reply is more personalized and intelligent, and the efficiency of automatic reply of the mail is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an email reply method, device, and electronic device. Background Art

[0002] As email becomes a primary tool for daily work and communication, individuals and businesses face the challenge of handling massive email volumes. Especially in corporate environments, employees spend significant time sifting, sorting, and responding to emails, which not only increases their workload but can also lead to missed important emails and delayed responses. Existing email auto-replies typically rely on fixed reply templates and simple rules.

[0003] However, automatic replying to emails based on fixed reply templates and simple rules results in poor accuracy and low intelligence of the content of the automatically replied emails, and requires users to spend a lot of time to adjust the reply. Summary of the Invention

[0004] The present invention provides an email reply method, device and electronic device, which are used to solve the defects of the prior art in that the automatic email reply content has poor accuracy, low intelligence and low reply efficiency.

[0005] The present invention provides an email reply method, comprising: Get the text of the email to be processed; Determine an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; Generate email reply content based on the email reply template and the intent recognition result of the email text to be processed; Perform an email reply based on the email reply content.

[0006] According to an email reply method provided by the present invention, generating email reply content based on the intent recognition result of the email reply template and the email text to be processed includes: Inputting the intent recognition result and the email reply template into a large language model to obtain an initial email reply output by the large language model; Get the historical email replies corresponding to the email type; Adjusting the initial email reply based on the personalized style of the historical email reply to obtain the email reply content; The personalized style includes user emotional style and / or user length style.

[0007] According to an email reply method provided by the present invention, replying to an email based on the email reply content includes: Obtaining email attribute information of the email text to be processed; Inputting the email text to be processed into a large language model to obtain an urgency value output by the large language model; Comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain a reply priority for the email text to be processed; Based on the reply priority, reply to the email reply content by email; The email attribute information includes at least one of sender information, email sending time, and historical reply records; The historical reply records are used to reflect the historical reply frequency and / or historical reply interval.

[0008] According to an email reply method provided by the present invention, the step of comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain the reply priority of the email text to be processed includes: determining a priority preference factor, wherein the priority preference factor is obtained based on historical priority adjustment information of the user; The reply priority is obtained by evaluating the email type, the email attribute information, the urgency value, and the priority preference factor.

[0009] According to an email reply method provided by the present invention, the step of obtaining the email type of the email text to be processed includes: Performing semantic analysis on the email text to be processed based on a large language model to obtain key email information output by the large language model; The email type is determined based on the key email information and the large language model.

[0010] According to an email reply method provided by the present invention, determining the email type based on the key email information and the large language model includes: determining a classification preference factor, wherein the classification preference factor is obtained based on historical type adjustment information of the user; The email type is determined based on the key email information, the classification preference factor, and the large language model.

[0011] According to an email reply method provided by the present invention, obtaining the email text to be processed includes: Obtaining a pending email, wherein the pending email includes at least one of an email subject, an email body, and an email attachment; Extracting text information of the email to be processed to obtain initial email text; The initial email text is pre-processed to obtain the email text to be processed.

[0012] According to an email reply method provided by the present invention, the file type of the email attachment includes a document type and / or an image type.

[0013] The present invention also provides an email replying device, comprising: Get unit to get the email text to be processed; A reply template determining unit, which determines an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; A reply content generating unit, which generates email reply content based on the email reply template and the intention recognition result of the email text to be processed; An email reply unit performs an email reply based on the email reply content.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-described email reply methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which implements any of the above-mentioned email reply methods when executed by a processor.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned email reply methods.

[0017] The email reply method, device and electronic device provided by the present invention determine the email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; based on the email reply template and the intention recognition result of the email text to be processed, generate email reply content, thereby improving the accuracy and professionalism of automatic email replies, and making the replies more personalized and intelligent, thereby greatly improving the efficiency of automatic email replies. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 1 is a flow chart of the email reply method provided by the present invention; Figure 2 It is a structural diagram of the email reply device provided by the present invention; Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0021] Traditional email classification methods rely primarily on simple keyword matching or rule-based engines. These methods often struggle to process complex email content, failing to fully understand the message and resulting in poorly accurate responses. Furthermore, existing automated responses are often based on fixed email templates and simple rules, resulting in limited accuracy and intelligence, and low response efficiency.

[0022] In response to the above problems, the present invention provides an email reply method to achieve accurate, intelligent, and efficient automatic email reply. Figure 1 Schematic diagram of the process of the email reply method provided by the present invention, such as Figure 1 As shown, the method includes: Step 110, obtaining the email text to be processed; Here, the "pending email text" refers to the text content of all unprocessed emails received in the user's mailbox. The pending email text can include the text content of multiple pending emails. Specifically, by accessing the user's mailbox, relevant data of the pending emails, such as the email title, email body, and email attachments, can be obtained. The obtained text information can then be de-noised to remove irrelevant data, and the email format can be standardized to provide standardized data input for subsequent processing.

[0023] Step 120: determining an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; Here, the email type refers to the classification of the email text to be processed, which can include multiple categories such as work emails, advertising emails, spam, social emails, etc. The email type can be used to reflect the importance and subject of the email text to be processed.

[0024] Furthermore, sentiment analysis results represent the emotional classification of the email text being processed, which can include positive, negative, interrogative, and requesting sentiments. These sentiment analysis results can be used to reflect the interaction needs of the email text being processed, helping to improve the accuracy of email responses. For example, for emails expressing gratitude or joy, positive and polite responses can be automatically generated; whereas for emails expressing negativity or complaints, sympathetic responses with proposed solutions can be generated.

[0025] Specifically, the text of any pending email can be fed into a large-scale pre-trained language model, such as BERT (Bidirectional Encoder Representations from Transformers). The language model then performs a deep semantic analysis of the email text, identifying the subject and key points of the email. The identified subject and key points are then used to determine the email type. For example, email types can include work emails, advertising emails, spam, and social media emails.

[0026] In addition, the text of any email to be processed can be input into the sentiment analysis model, and the sentiment features of the email text to be processed can be extracted by the sentiment analysis model. The sentiment features can be classified and predicted to obtain the sentiment analysis results of the email text to be processed. Such as positive, negative, question, request, etc., thereby realizing the sentiment classification of the email text to be processed. It should be noted that the sentiment analysis model here can be constructed based on a deep learning algorithm, for example, it can be obtained by training an initial neural network model, or by fine-tuning a general large-scale language model, or directly using a general large-scale language model. It should be noted that the embodiment of the present invention does not specifically limit the order of obtaining the email type of the email text to be processed and the sentiment analysis results.

[0027] Then, an email reply template that matches the email type and sentiment analysis results can be retrieved from the reply template library. For example, a target email reply template that matches the email type can be retrieved from the reply template library. Then, an email reply template that matches the sentiment analysis results can be retrieved from the target email reply template, thereby obtaining an email reply template that corresponds to the email type and sentiment analysis results. It should be noted that it is also possible to first retrieve a target email reply template that matches the sentiment analysis results from the reply template library, and then retrieve an email reply template that matches the email type from the target email reply template.

[0028] It can be understood that by determining an email reply template that matches the email type and sentiment analysis results of the email text to be processed, the subject and sentiment of the final generated email reply content can be consistent with the email text to be processed, greatly improving the accuracy of the generated email reply content.

[0029] Step 130: Generate email reply content based on the email reply template and the intent recognition result of the email text to be processed; Specifically, the email reply template, the intent recognition result of the email text to be processed, and the reply prompt text can be input into the large language model. The large language model can generate the intent reply content for the intent recognition result according to the instructions of the reply prompt text, and fill the intent reply into the email reply template. For example, if the email text to be processed contains the expression "schedule a meeting" or "get a quote", these intentions can be automatically identified, and more accurate reply content can be generated. It can be understood that the intent reply content generated here contains the text elements that are missing in the email reply template.

[0030] It's important to note that sentiment analysis and intent recognition technologies enable a comprehensive understanding of the emotional tone (e.g., positive, negative, request) and core intent (e.g., scheduling a meeting, requesting information) within the email text being processed. This allows for the generation of appropriate responses based on the tone and content of the email. The introduction of sentiment analysis ensures that the tone and expression of the generated email responses are more consistent with the emotional orientation of the email, while intent recognition accurately captures the core needs of the email text being processed. This not only improves the accuracy and professionalism of automated email replies, but also makes them more personalized and intelligent, enhancing the user experience and preventing users from wasting time adjusting automatically generated email replies, significantly improving the efficiency of automated email replies.

[0031] Step 140: Reply via email based on the email reply content.

[0032] Specifically, after obtaining the email reply content of any pending email text, the pending email text can be automatically replied to by the email reply content. Alternatively, the email reply content can be replied to by email according to the reply priority order of the pending email text.

[0033] The method provided by the embodiment of the present invention determines an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; generates email reply content based on the email reply template and the intention recognition result of the email text to be processed, thereby improving the accuracy and professionalism of automatic email replies, and making the replies more personalized and intelligent, thereby greatly improving the efficiency of automatic email replies.

[0034] Based on any of the above embodiments, step 130 includes: Inputting the intent recognition result and the email reply template into a large language model to obtain an initial email reply output by the large language model; Get the historical email replies corresponding to the email type; Adjusting the initial email reply based on the personalized style of the historical email reply to obtain the email reply content; The personalized style includes user emotional style and / or user length style.

[0035] Specifically, first, the intent recognition result and the email reply template are combined to construct a reply prompt text. Then, the reply prompt text is input into a large language model, which then outputs an initial email reply according to the requirements of the reply prompt text. It is understood that the initial email reply here can be considered as a reply content that conforms to the email type in terms of format, conforms to the intent recognition result in terms of content, and conforms to the sentiment analysis result in terms of emotion.

[0036] Furthermore, to ensure that email replies are tailored to the user's response style, the recipient's historical email replies for this email type can be obtained. This historical email reply can include multiple historical emails. It should be noted that this historical email reply can include a record of adjustments made by the user to the automatically generated email collection reply for that round. This historical email reply reflects the user's historical behavior, feedback data, and email processing habits, and thus can reflect the user's response preferences.

[0037] Next, the personalized style of historical email replies can be extracted. Using the personalized style, the initial email reply can be adjusted to obtain the email reply content. Specifically, the personalized style here includes the user's emotional style and / or the user's length style. The user's emotional style is used to reflect the user's personalized reply tone. For example, high-frequency modal particles in historical email replies can be extracted as the user's emotional style. Furthermore, the user's length style is used to reflect the user's preferred email length (number of words, number of paragraphs) and information density, such as whether detailed background information is included. This can be obtained by counting the average sentence length, number of paragraphs, and total number of words in historical email replies.

[0038] Finally, the emotion and / or length of the initial email reply can be adjusted based on the user's emotional style and / or user's length style of historical email replies, so that the final email reply content conforms to the user's emotional style and / or user's length style. For example, the user's commonly used modal particles can be added to the initial email reply. Alternatively, the length of the initial email reply can be adjusted according to the user's length style. For example, some users prefer concise replies, while other users may prefer more detailed explanations. Therefore, adjusting the initial email reply based on the user's emotional style and / or user's length style not only improves the efficiency of email processing, but also greatly reduces the need for manual intervention.

[0039] The method provided by the embodiments of the present invention uses an online learning mechanism to continuously optimize email reply content based on historical email replies that reflect users' historical behavior, feedback data, and email processing habits, ensuring that each email processing is tailored to the user's actual needs and personalized requirements. It is understood that over time, the processing results of the email reply method provided by the embodiments of the present invention will become increasingly accurate, providing more efficient and intelligent services.

[0040] To further enhance the intelligence of the automatic email reply, based on any of the above embodiments, step 140 includes: Obtaining email attribute information of the email text to be processed; Inputting the email text to be processed into a large language model to obtain an urgency value output by the large language model; Comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain a reply priority for the email text to be processed; Based on the reply priority, reply to the email reply content by email; The email attribute information includes at least one of sender information, email sending time, and historical reply records; The historical reply records are used to reflect the historical reply frequency and / or historical reply interval.

[0041] Specifically, first, obtain the email attribute information of the email text to be processed. The email attribute information here may include at least one of sender information, email sending time, and historical reply records. The historical reply records are used to reflect the historical reply frequency and / or historical reply interval.

[0042] Here, the historical reply frequency can reflect the level of active communication between the user and the sender. It's understandable that a higher historical reply frequency indicates frequent communication between the two parties, and timely replies may be necessary to maintain communication continuity, thus giving the pending email a higher reply priority. Furthermore, the historical reply interval can reflect the sender's expected time for an email reply. It's understandable that a shorter historical reply interval indicates that the sender likely desires a timely reply, thus giving the pending email a higher reply priority.

[0043] Next, the email text to be processed can be input into the large language model to obtain the urgency value output by the large language model. It is understood that the urgency value here can be used to reflect the urgency of the content of the email text to be processed. For example, if the email text to be processed contains "Please process as soon as possible", the urgency value of the email text to be processed is high.

[0044] Furthermore, the response priority of the pending email text can be evaluated by comprehensively considering the email type, email attribute information, and urgency value. Specifically, different initial priority scores can be assigned to different email types based on actual business needs. For example, customer complaint emails and emergency notification emails may have higher initial priority scores, followed by work emails, and advertising emails may have the lowest initial priority score. Furthermore, based on the email attribute information, an email attribute score can be determined that matches the sender information, email sending time, or historical reply records. For example, a higher email attribute score may be assigned if the sender information indicates that the sender is an important customer; a higher email attribute score may be assigned if the email was sent during business hours; and a higher email attribute score may be assigned if historical reply records indicate a high frequency of historical replies and / or short intervals between historical replies. A weighted summation method can then be used to calculate the final priority score by combining the weights and scores corresponding to the email type, email attribute information, and urgency value. This priority score can then be used as the reply priority for the pending email text.

[0045] Alternatively, more complex machine learning models (such as decision trees and neural networks) can be used to build an evaluation model. By learning from large amounts of historical email data, the evaluation model can automatically determine the weights of various factors and the evaluation rules. In practical applications, the evaluation model can be fed with email type, email attribute information, and urgency values, and the model can output the response priority for the pending email text.

[0046] Finally, you can sort the email replies according to their priority. The email replies with higher priority will be replied to first.

[0047] It's important to note that emails are automatically prioritized by combining factors such as the text of the pending email, sender information, past interactions, and the email's urgency. Using reinforcement learning algorithms, the email prioritization strategy is dynamically adjusted based on user habits and feedback, ensuring that high-priority emails are addressed first. This significantly improves email processing efficiency, making it particularly useful in busy work environments, helping users respond to critical emails promptly and preventing them from being overlooked.

[0048] The method provided in the embodiment of the present invention evaluates the reply priority of the email text to be processed by comprehensively considering the email type, email attribute information and urgency value, obtains the priority division that meets the actual needs of the user, and replies to the email reply content according to the accurate reply priority, thereby greatly improving the intelligence of the automatic email reply and thus greatly improving the user experience.

[0049] In order to continuously optimize the accuracy of the reply priority of the pending email, based on any of the above embodiments, the step of comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain the reply priority of the pending email text includes: determining a priority preference factor, wherein the priority preference factor is obtained based on historical priority adjustment information of the user; The reply priority is obtained by evaluating the email type, the email attribute information, the urgency value, and the priority preference factor.

[0050] Specifically, historical priority adjustment information can be obtained by obtaining the user's priority adjustment records for historical email replies within a preset time period. The priority adjustment records herein can be raising or lowering the priority of historical email replies. It is understood that historical priority adjustment information can be used to reflect the user's real-time demand for email reply priority.

[0051] Then, the priority preference factor can be determined based on the historical priority adjustment information. For example, when the historical priority adjustment information indicates an increase in priority, the priority preference factor is increased accordingly; when the historical priority adjustment information indicates a decrease in priority, the priority preference factor is decreased accordingly.

[0052] Furthermore, the reply priority can be evaluated by comprehensively considering the email type, email attribute information, urgency value, and priority preference factor. The reply priority can be calculated using a weighted summation method, or by inputting the email type, email attribute information, urgency value, and priority preference factor into a machine learning model (such as a decision tree or neural network) to determine the reply priority. It should be noted that when outputting the reply priority using a machine learning model, a reinforcement learning algorithm can be employed to output the reply priority, ensuring that each sorting process better meets user needs.

[0053] The method provided by an embodiment of the present invention determines a priority preference factor through the user's historical priority adjustment information, and takes the priority preference factor into consideration in the process of determining the current reply priority, so that the current reply priority can better meet the user's current actual needs, thereby providing personalized email management services.

[0054] Based on any of the above embodiments, the step of obtaining the email type of the email text to be processed includes: Performing semantic analysis on the email text to be processed based on a large language model to obtain key email information output by the large language model; The email type is determined based on the key email information and the large language model.

[0055] Specifically, the email text to be processed can be input into a large language model, which performs semantic analysis on the email text and outputs key email information. It should be noted that this key email information can include the email subject and summary. The key email information can then be further input into the large language model, which performs classification prediction based on the key email information to determine the email type.

[0056] It should be noted that, compared to existing techniques that determine the email type of an email text based on preset matching rules, the present invention uses a large-scale language model to perform semantic analysis on the email text to obtain key email information. The email type is then determined based on this key information and the large-scale language model, significantly improving the accuracy and efficiency of email text classification.

[0057] The method provided by the embodiments of this invention utilizes a large, pre-trained language model to conduct in-depth analysis of email content, accurately identifying the subject, key information, and context of an email, thereby achieving precise email classification. This not only addresses the issues inherent in traditional email classification methods, which rely on keyword matching and rule engines, but also enables the processing of more complex and diverse emails, significantly improving email classification accuracy and processing efficiency. Whether it's work email, advertising email, or spam, it can be quickly and accurately categorized, reducing manual intervention and screening time.

[0058] To further improve the accuracy of the email type, based on any of the above embodiments, determining the email type based on the key email information and the large language model includes: determining a classification preference factor, wherein the classification preference factor is obtained based on historical type adjustment information of the user; The email type is determined based on the key email information, the classification preference factor, and the large language model.

[0059] Specifically, historical priority adjustment information can be obtained by obtaining a record of the type adjustment of historical email replies made by a user within a preset time period. The historical type adjustment information here can include adjusting the type of a historical email reply from advertising email to spam. It is understood that historical priority adjustment information can be used to reflect a user's real-time needs for email classification.

[0060] Then, the classification preference factor can be determined based on the user's historical type adjustment information. For example, if a user adjusts the type of a historical email reply from advertising email to spam, the weight of the advertising email type can be increased when predicting email classification.

[0061] Furthermore, by inputting key email information and classification preference factors into a large language model, classification prediction can be performed based on the key email information through the large language model. During the prediction process, the classification preference factors can be combined to adjust the initial prediction results to output the final email type.

[0062] The method provided by the embodiment of the present invention continuously optimizes the email type of the current email text to be processed by combining the feedback information of historical email replies when classifying the email text to be processed, thereby improving the accuracy of the output email type and making it more in line with the real-time needs of users.

[0063] Based on any of the above embodiments, step 110 includes: Obtaining a pending email, wherein the pending email includes at least one of an email subject, an email body, and an email attachment; Extracting text information of the email to be processed to obtain initial email text; The initial email text is pre-processed to obtain the email text to be processed.

[0064] Based on any of the above embodiments, the file type of the email attachment includes a document type and / or an image type.

[0065] Specifically, it is possible to obtain the pending emails received in the user's mailbox. The pending emails here include at least one of the email subject, email body, and email attachments. It should be noted that the file type of the email attachment here includes document type and / or image type. The document type here may include docx, doc, PDF, Excel. For example, if the email attachment of the pending email is a contract or agreement, the text information in the attachment can be extracted and correlated with the content of the email body for analysis, so that the attachment information can be taken into account when classifying emails, prioritizing, and generating replies.

[0066] Then, you can extract the text information of the email to be processed to obtain the initial email text. For the email subject and body, you can directly obtain the text corresponding to the email subject and body to obtain the text information of the email to be processed. For email attachments of docx, doc, and Excel types, you can use document parsing tools to extract the text information from document-type email attachments. Furthermore, for PDF and image-type email attachments, you can use OCR (Optical Character Recognition) technology to extract text information. Furthermore, for image-type email attachments, you can also use image recognition technology to extract text information from the image.

[0067] The extracted text information from the emails to be processed can then be spliced ​​together to produce the initial email text. It's important to note that these multimodal technologies not only process traditional text emails but also comprehensively analyze all information within the email and its attachments, ensuring that no critical information is missed. Extracting text information from email attachments enables a more comprehensive understanding of the email content and accurate processing, even for complex emails (such as those containing tables, images, or signatures).

[0068] Furthermore, the initial email text can be preprocessed to obtain the email text to be processed. This preprocessing can include removing advertisements, marketing information, irrelevant text, or repetitive content from the initial email text to ensure high-quality text data for subsequent processing. Furthermore, the initial email text undergoes natural language processing operations such as word segmentation and stop word removal to standardize the initial email text and facilitate subsequent semantic analysis and classification.

[0069] It is understandable that for some emails with more complex formats, important content can be extracted and organized into structured data to ensure that the email information can be accurately delivered to subsequent processing procedures.

[0070] The method provided by the embodiments of the present invention not only recognizes the email body but also fully understands the entire email content, including key text or images in attachments. This multimodal processing capability enables comprehensive analysis and efficient processing of all relevant information in complex emails, further improving the accuracy and reliability of email processing.

[0071] Based on any of the above embodiments, the present invention further provides an automatic email reply system, the system comprising: Email data input and preprocessing module: This module accesses the user's mailbox through an interface, obtains relevant email data (such as title, body, attachments, sender information, etc.), denoises the email, removes irrelevant data, and standardizes the email format to provide standardized data input for subsequent processing.

[0072] Email Classification and Prioritization Module: Based on large-scale pre-trained language models such as BERT, this module conducts in-depth analysis of email content and automatically identifies email categories (e.g., work emails, advertising emails, spam, etc.). Furthermore, it uses reinforcement learning algorithms to prioritize emails based on multiple factors, including email content, sender information, past email interactions, and email time, ensuring that high-priority emails are processed first.

[0073] Sentiment Analysis and Intent Identification Module: This module combines sentiment analysis and natural language understanding technologies to automatically identify the emotional tendencies (e.g., positive, negative, inquiring, etc.) and core intent (e.g., requesting information, scheduling a meeting, confirming a matter, etc.) in emails. Sentiment analysis helps the system adjust the tone of responses, while intent identification guides the subsequent automatic response generation.

[0074] It's important to note that by combining intelligent processing with sentiment and intent analysis, email replies go beyond simple automatic text generation and instead prioritize in-depth understanding and precise feedback. This intelligent email management approach not only improves email processing efficiency but also provides a more humane and personalized service, significantly enhancing the user experience.

[0075] Automatic reply generation module: Based on email classification, sentiment analysis, and intent recognition, this module automatically generates email replies. By leveraging a large-scale pre-trained language model, the system generates personalized, accurate automatic replies based on the email's content and sentiment. The tone and content are automatically adjusted based on the email's type and urgency.

[0076] Personalized learning and optimization module: This module continuously learns based on the user's email processing history and interactive behavior, automatically adjusts email classification, sorting and reply strategies to continuously optimize the system's email processing results and provide personalized email management services.

[0077] Multimodal attachment processing module: In addition to text, the system also supports analysis of attachment content. Using OCR and image recognition technology, the system can extract important information from attachments and combine it with the email body for unified classification and reply generation, further improving the comprehensiveness of email processing.

[0078] The automated email reply system provided by this invention effectively improves email classification accuracy, automatically prioritizes emails, reduces email processing time and manual intervention, and generates precise automated replies based on the sentiment and intent of the email content, thereby improving work efficiency and user experience. Through a deep learning approach based on large models and a continuously optimized learning mechanism, it provides an efficient, intelligent, and personalized email processing solution suitable for a variety of fields, including personal mailbox management, enterprise email systems, and customer service.

[0079] Based on any of the above embodiments, Figure 2 Schematic diagram of the structure of the mail reply device provided by the present invention, such as Figure 2 As shown, the device includes: An acquiring unit 210 acquires the email text to be processed; The reply template determining unit 220 determines an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; A reply content generating unit 230 generates email reply content based on the email reply template and the intention recognition result of the email text to be processed; The email reply unit 240 performs an email reply based on the email reply content.

[0080] The device provided by the embodiment of the present invention determines an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; generates email reply content based on the email reply template and the intention recognition result of the email text to be processed, thereby improving the accuracy and professionalism of automatic email replies, and making the replies more personalized and intelligent, thereby greatly improving the efficiency of automatic email replies.

[0081] Based on any of the above embodiments, the reply content generating unit is specifically configured to: Inputting the intent recognition result and the email reply template into a large language model to obtain an initial email reply output by the large language model; Get the historical email replies corresponding to the email type; Adjusting the initial email reply based on the personalized style of the historical email reply to obtain the email reply content; The personalized style includes user emotional style and / or user length style.

[0082] Based on any of the above embodiments, the email reply unit is specifically configured to: Obtaining email attribute information of the email text to be processed; Inputting the email text to be processed into a large language model to obtain an urgency value output by the large language model; Comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain a reply priority for the email text to be processed; Based on the reply priority, reply to the email reply content by email; The email attribute information includes at least one of sender information, email sending time, and historical reply records; The historical reply records are used to reflect the historical reply frequency and / or historical reply interval.

[0083] Based on any of the above embodiments, the email reply unit is further specifically configured to: determining a priority preference factor, wherein the priority preference factor is obtained based on historical priority adjustment information of the user; The reply priority is obtained by evaluating the email type, the email attribute information, the urgency value, and the priority preference factor.

[0084] Based on any of the above embodiments, the reply template determining unit is specifically configured to: Performing semantic analysis on the email text to be processed based on a large language model to obtain key email information output by the large language model; The email type is determined based on the key email information and the large language model.

[0085] Based on any of the above embodiments, the reply template determining unit is further specifically configured to: determining a classification preference factor, wherein the classification preference factor is obtained based on historical type adjustment information of the user; The email type is determined based on the key email information, the classification preference factor, and the large language model.

[0086] Based on any of the above embodiments, the acquiring unit is specifically configured to: Obtaining a pending email, wherein the pending email includes at least one of an email subject, an email body, and an email attachment; Extracting text information of the email to be processed to obtain initial email text; The initial email text is pre-processed to obtain the email text to be processed.

[0087] Based on any of the above embodiments, the file type of the email attachment includes a document type and / or an image type.

[0088] Figure 3 An example of a physical structure diagram of an electronic device is shown below. Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other via the communications bus 340. The processor 310 may invoke logic instructions in the memory 330 to execute an email reply method, which includes: obtaining an email text to be processed; determining an email reply template corresponding to the email type and sentiment analysis results of the email text to be processed; generating email reply content based on the email reply template and the intent recognition results of the email text to be processed; and replying to the email based on the email reply content.

[0089] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0090] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the email reply method provided by the above methods, which includes: obtaining the email text to be processed; determining an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; generating email reply content based on the email reply template and the intention recognition result of the email text to be processed; and replying to the email based on the email reply content.

[0091] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the email reply method provided by the above-mentioned methods, the method comprising: obtaining the email text to be processed; determining an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; generating email reply content based on the email reply template and the intention recognition result of the email text to be processed; and performing an email reply based on the email reply content.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0093] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An email reply method, characterized in that: include: Get the text of the email to be processed; Determine an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; Generate email reply content based on the email reply template and the intent recognition result of the email text to be processed; Perform an email reply based on the email reply content.

2. The email reply method according to claim 1, characterized in that: The generating of email reply content based on the intention recognition result of the email reply template and the email text to be processed includes: Inputting the intent recognition result and the email reply template into a large language model to obtain an initial email reply output by the large language model; Get the historical email replies corresponding to the email type; Adjusting the initial email reply based on the personalized style of the historical email reply to obtain the email reply content; The personalized style includes user emotional style and / or user length style.

3. The email reply method according to claim 1, wherein: The step of replying to an email based on the email reply content includes: Obtaining email attribute information of the email text to be processed; Inputting the email text to be processed into a large language model to obtain an urgency value output by the large language model; Comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain a reply priority for the email text to be processed; Based on the reply priority, reply to the email reply content by email; The email attribute information includes at least one of sender information, email sending time, and historical reply records; The historical reply records are used to reflect the historical reply frequency and / or historical reply interval.

4. The email reply method according to claim 3, wherein: The step of comprehensively evaluating the email type, the email attribute information, and the urgency value to obtain a reply priority of the email text to be processed includes: determining a priority preference factor, wherein the priority preference factor is obtained based on historical priority adjustment information of the user; The reply priority is obtained by evaluating the email type, the email attribute information, the urgency value, and the priority preference factor.

5. The email reply method according to any one of claims 1 to 4, characterized in that: The step of obtaining the email type of the email text to be processed includes: Performing semantic analysis on the email text to be processed based on a large language model to obtain key email information output by the large language model; The email type is determined based on the key email information and the large language model.

6. The email reply method according to claim 5, characterized in that: The determining the email type based on the key email information and the large language model includes: determining a classification preference factor, wherein the classification preference factor is obtained based on historical type adjustment information of the user; The email type is determined based on the key email information, the classification preference factor, and the large language model.

7. The email reply method according to any one of claims 1 to 4, characterized in that: The step of obtaining the email text to be processed includes: Obtaining a pending email, wherein the pending email includes at least one of an email subject, an email body, and an email attachment; Extracting text information of the email to be processed to obtain initial email text; The initial email text is pre-processed to obtain the email text to be processed.

8. The email reply method according to claim 7, characterized in that: The file type of the email attachment includes a document type and / or an image type.

9. An email reply device, characterized in that: include: Get unit to get the email text to be processed; A reply template determining unit, which determines an email reply template corresponding to the email type and sentiment analysis result of the email text to be processed; A reply content generating unit, which generates email reply content based on the email reply template and the intention recognition result of the email text to be processed; The email reply unit performs an email reply based on the email reply content.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the email reply method according to any one of claims 1 to 8 is implemented.

Images