Event feedback filling method and system, terminal device and readable storage medium

Abstract

This invention provides an event feedback filling method and system, a terminal device, and a readable storage medium. The event feedback filling method includes: acquiring voice information input by a user based on an event feedback page and converting it into corresponding text information; extracting key elements from the text information based on an element extraction model; combining the text information and the key elements extracted by the element extraction model to perform information fusion and summarization of the event handling situation, generating feedback text; and filling the generated feedback text into the feedback content input box of the event feedback page to complete the automatic filling of the event feedback, which can effectively improve the accuracy and completeness of feedback information extraction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an event feedback filling method and system, a terminal device, and a readable storage medium. Background Technology

[0002] In mobile applications, the incident feedback function is a key module for protecting user rights and improving service quality. Traditional incident feedback pages often rely on plain text input or form filling, requiring users to manually enter detailed information such as ID number, incident address, incident level, and processing result. This approach has significant limitations: firstly, manual input is cumbersome, especially for users who are not good at typing or are in mobile environments, which is time-consuming and laborious, easily reducing their willingness to provide feedback; secondly, the information entered by users is prone to being formatted incorrectly or missing key elements, requiring staff to spend a lot of time organizing, verifying, and supplementing it, seriously affecting the efficiency of incident processing.

[0003] With the development of speech recognition technology, some apps have attempted to add voice input functionality to their feedback pages, allowing users to describe events by voice. However, most existing solutions can only achieve basic speech-to-text conversion, resulting in a large amount of irrelevant content mixed in with the text (such as interjections, filler words, and background noise), affecting the accuracy of information extraction. In addition, users still need to manually check and supplement the information, failing to fundamentally improve feedback efficiency and still resulting in incomplete feedback information, which affects the subsequent processing of events. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides an event feedback filling method and system, a terminal device, and a readable storage medium, thereby improving the accuracy and completeness of feedback information extraction.

[0005] The technical solution provided by this invention is as follows: On one hand, the present invention provides an event feedback filling method, comprising: Obtain the voice information input by the user based on the event feedback page and convert it into corresponding text information; Key elements are extracted from the text information based on the element extraction model. By combining the text information and the key elements extracted by the element extraction model, information fusion and summarization of the event handling situation are performed to generate feedback text. The generated feedback text is then filled into the feedback content input box on the event feedback page to complete the automatic filling of the event feedback.

[0006] In some preferred embodiments, the step of extracting elements from the text information based on the element extraction model includes: The user's ID card number is identified based on pre-configured regular expression matching. Identify place names in the text information and obtain the address where the event occurred based on a pre-configured place name database; Identify keywords related to event levels in the text information and obtain the event level based on a pre-configured event level keyword library; Extract keywords related to the event processing result from the text information, and obtain the event processing result by combining the context of the keywords.

[0007] In some preferred embodiments, by combining the text information and the key elements extracted by the element extraction model, information fusion and summarization of the event handling situation are performed to generate feedback text, including: A summarized context vector is generated based on the text information and the key elements extracted by the element extraction model. Based on pre-configured constraint optimization objectives, a conditional language model is used to generate an output sequence word by word to obtain the feedback text. The constraint optimization objectives include the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and a preset standard template.

[0008] In some preferred embodiments, after filling the feedback content input box of the event feedback page with the generated feedback text, the method further includes: Based on the pre-set list of necessary elements for event feedback, confirm whether the key elements extracted by the element extraction model are complete; To ensure all information is complete, a submit button should be provided on the event feedback page for users to confirm their submission. If an element is incomplete, a missing element prompt will be displayed on the event feedback page for the user to supplement. After the user supplements the missing element, the user will be prompted to re-enter the step to confirm whether the key element is complete.

[0009] On the other hand, the present invention provides an event feedback filling system, comprising: The voice input module is used to acquire voice information input by the user based on the event feedback page; A speech-to-text module, connected to the speech input module, is used to convert the speech information acquired by the speech input module into corresponding text information; The element extraction module, connected to the speech-to-text module, is used to extract key elements from the text information based on the element extraction model. The text summarization module, connected to the element extraction module, is used to combine the text information and the key elements extracted by the element extraction model to perform information fusion and summarization on the event handling situation and generate feedback text. The text filling module, connected to the text summarization module, is used to fill the generated feedback text into the feedback content input box of the event feedback page, thereby completing the automatic filling of the event feedback.

[0010] In some preferred embodiments, the feature extraction module includes: The ID card number recognition unit is used to identify the user's ID card number based on a pre-configured regular expression. An event address identification unit is used to identify place names in the text information and obtain the event location based on a pre-configured place name database. An event level identification unit is used to identify keywords related to event levels in the text information and obtain the event level based on a pre-configured event level keyword library; The event processing result recognition unit is used to extract keywords related to the event processing result from the text information and obtain the event processing result by combining the context of the keywords.

[0011] In some preferred embodiments, the text summarization module includes: The context vector induction unit is used to generate an inductive context vector based on the text information and the key elements extracted by the element extraction model. The feedback text generation unit, connected to the context vector induction unit, is used to generate an output sequence word by word using a conditional language model based on a pre-configured constraint optimization objective, thereby obtaining the feedback text. The constraint optimization objective includes the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and a preset standard template.

[0012] In some preferred embodiments, the event feedback population system further includes: The element verification module, connected to the element extraction module, is used to confirm whether the key elements extracted by the element extraction model are complete based on a list of necessary elements fed back by a preset event. The submission module is connected to the element verification module. If the element verification module confirms that all elements are complete, it provides a submission button on the event feedback page for the user to confirm the submission. The manual supplementation module is connected to the element verification module. If the element verification module confirms that the elements are incomplete, it displays a missing element prompt on the event feedback page for the user to supplement, and sends the supplemented key elements to the element verification module after the user supplements.

[0013] In another aspect, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described event feedback filling method.

[0014] In another aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described event feedback filling method.

[0015] The event feedback filling method and system, terminal device and readable storage medium provided by the present invention can bring at least the following beneficial effects: 1) Improve the accuracy of information extraction: The configured text noise reduction module removes irrelevant information and invalid text from the initial text information, providing clean text data for subsequent feature extraction, effectively improving the accuracy of key feature extraction and reducing key feature identification errors caused by text noise; 2) Improve event feedback efficiency: By using the element extraction module and text summarization module, key elements are automatically extracted and event handling status is automatically summarized, eliminating the need for users to manually input and organize information, thereby significantly shortening the user feedback operation time and improving feedback efficiency; at the same time, by automatically filling in feedback text, the number of user operation steps is reduced, improving the user feedback experience and willingness. 3) Ensure the completeness of feedback information: The element verification module automatically determines whether the feedback elements are complete and promptly reminds users to supplement missing elements, avoiding incomplete feedback information due to user omissions, providing complete and effective data support for subsequent event processing, and reducing the workload of staff in verifying information; 4) Enhanced system applicability: Each module of the system has a certain degree of flexibility, and the necessary element list, noise reduction rule base and element extraction model can be adjusted according to the business needs of different APPs. It is suitable for various APPs that require event feedback functions and has a wide range of applications. 5) Lowering the barrier to user operation: By combining voice input with automatic processing, users do not need to have proficient text input skills, which is especially suitable for the elderly, mobile users and other groups, thus lowering the barrier to user operation and improving the universality of the system. Attached Figure Description

[0016] The preferred embodiments will now be described in a clear and easy-to-understand manner, with reference to the accompanying drawings, to further explain the above-mentioned characteristics, technical features, advantages, and implementation methods.

[0017] Figure 1 This is a schematic flowchart of one embodiment of the event feedback filling method in this invention; Figure 2 This is a schematic flowchart of another embodiment of the event feedback filling method in this invention; Figure 3 This is a schematic flowchart of one embodiment of the event feedback filling system of the present invention; Figure 4 This is a schematic diagram of the terminal device structure in this invention.

[0018] Figure label: 100 - Event feedback filling system, 110 - Voice input module, 120 - Voice to text module, 130 - Feature extraction module, 140 - Text summarization module, 150 - Text filling module. Detailed Implementation

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0020] A first embodiment of the present invention provides an event feedback filling method, such as... Figure 1 As shown, it includes: S10 acquires the voice information input by the user based on the event feedback page and converts it into corresponding text information; S20 Extracts key elements from text information based on an element extraction model; S30 combines textual information and key elements extracted by the element extraction model to perform information fusion and summarization on the event handling situation and generate feedback text. S40 fills the generated feedback text into the feedback content input box on the event feedback page, completing the automatic filling of the event feedback.

[0021] In this embodiment, the event feedback page is a page in an application app that requires users to input feedback information about events. For example, it could be a page in an enterprise service app where employees report repair issues, or a page in a campus service app where students report issues related to the cafeteria. The application app can also be a government service app, a lifestyle service app, etc. To lower the barrier to entry for users, a voice input option is provided on the event feedback page, allowing users to record voice information about the event handling process using the built-in microphone of their terminal device, and then convert the recorded voice information into text information. The method for converting voice information into text information is not specifically limited here; any existing speech recognition algorithm can be used, such as a deep learning-based speech recognition model, to perform text transcription processing on the received voice information, converting it into the corresponding initial text information.

[0022] After obtaining the initial text information, it is input into a pre-installed and trained feature extraction model to extract key features. Semantic analysis and feature extraction are performed on the text information, including: identifying the user's ID number based on pre-configured regular expression matching; identifying place names in the text information and obtaining the event location based on a pre-configured place name database; identifying keywords related to the event level in the text information and obtaining the event level based on a pre-configured event level keyword database; and extracting keywords related to the event processing result from the text information and combining the context of the keywords to obtain the event processing result. Specifically: The process of identifying a user's ID card number is as follows: A regular expression for an 18-digit ID card number is predefined. After receiving text information, all consecutive strings are searched within the text, and each string is checked to see if it can be completely matched by the defined regular expression. Finally, the string containing all strings in the text that conform to the ID card number format is returned, completing the matching of the user's ID card number.

[0023] The process of identifying the address where the event occurred is as follows: A place name database is pre-built, where each record is a structured standard address containing: province, city, district, street, and number. Upon receiving text information, a pre-built named entity recognition model is used to perform a preliminary scan of the text information, identifying and extracting all possible place name entity fragments (corresponding to the place names mentioned above), forming a list of place name entities. For example, from the text "I went to Zhongguancun Street, Haidian District, Beijing," a list of place name entities containing "Beijing," "Haidian District," and "Zhongguancun Street" may be identified. Next, a pre-configured similarity function is used to score the identified list of place name entities, finally outputting a standardized address vector containing the province, city, district, street, and number. It should be understood that the information contained in the mentioned standard address can be determined according to the actual situation; the above is just an example of one approach.

[0024] To improve the accuracy of event location extraction, similarity ( A , g Based on word overlap J ( A ∩name( g )) and hierarchical consistency L hier( A , g Together, we determine and finally extract the standard address with the highest similarity score for output: similarity A , g )= α • J ( A ∩name( g ))+ β • L hier( A , g ) in, A Represents a list of identified place-name entities; g This represents the standard address in the constructed place name database G; α and β Representing word overlap J ( A ∩name( g )) and hierarchical consistency L hier( A , g The weighting coefficients of ) and α + β =1, the specific value is determined based on the actual situation. Word overlap. J ( A ∩name( g This is used to measure the degree of textual overlap between place names identified in textual information and place names contained in standard addresses. J ( A ∩name( g ))=| A ∩name( g )| / A ∪name( g For example, the user-input text includes: Zhongguancun Street, Haidian District, Beijing; there is a record in the place name database: g = No. 27, Zhongguancun Street, Haidian District, Beijing; according to the standards of province, city, district, street, and number, the intersection is... A ∩name(g) = {Beijing, Beijing, Haidian, Zhongguancun Street}, Union A ∪name(g)={Beijing, Beijing, Haidian, Zhongguancun Street, No. 27}, J( A ∩name( g =0.8. Hierarchical consistency L hier( A , g This is used to assess the degree of match / conflict between identified place names and candidate standard addresses in terms of administrative hierarchy. For example, if the user inputs text information including: Haidian District, Beijing; and there is a record in the place name database g = Chaoyang District, Beijing, according to the standard of province, city, district, street, and number, the two pieces of information are respectively (Beijing, Beijing, Haidian) and (Beijing, Beijing, Chaoyang), then... L hier( A , g =2 / 3.

[0025] The process of identifying event levels is as follows: A set of keywords is predefined for each event level, for example: emergency level event. K urgent = {immediate, right now, at the last minute}; matters of high importance. K high = {important, significant, serious, critical}; general level of event. K medium = {normal, common, regular}; minor level event K `low` is defined as {minor, minor issue, convenient, when available}, etc. After receiving the text message, the matching strength between it and each event level is calculated. Specifically, when calculating the matching strength with emergency level events, the emergency level events appearing in the text message are first counted. K The number of words in the urgent dictionary, then divided by the urgency level of the event. K The total number of words in the urgent dictionary, resulting in a value between 0 and 1, represents the intensity of the text's mention of that level. This process continues until the event level with the highest score is output. For example, if the text contains the word "urgent," the event level is urgent. K The urgent dictionary contains 4 words, so the matching strength S_level = 1 / 5 = 0.2.

[0026] To avoid situations where multiple event levels with the highest matching scores appear simultaneously, making it impossible to make a judgment, this embodiment further introduces context embedding vectors for disambiguation. Specifically: First, the entire text information is input into the BERT model or other similar models to obtain a high-dimensional vector representing its overall semantics. Then, the average semantic vector for each event level is obtained; for example, for emergency level events... KFor urgent events, the average semantic vector for that level is obtained by averaging the BERT vectors of all words in the dictionary. Next, the cosine similarity between the high-dimensional vector of the text information and the average semantic vector of each event level is calculated. Finally, the closest event level is determined based on the calculated cosine similarity.

[0027] The process for identifying processing results is as follows: A corresponding keyword set is predefined for each processing result, for example: resolved status Kresolved = {resolved, completed}; processing status Kprocessing = {processing, accepted}; requiring coordination status Kpending = {requiring coordination, pending follow-up, further coordination needed}, etc. After receiving text information, the number of keywords appearing in the corresponding status keyword set is counted, and finally, the text information is assigned to the status result category with the most frequently appearing keywords. In other embodiments, lightweight classifiers such as softmax can also be used for classification.

[0028] After extracting the key elements, these elements are sent to a text induction model to fuse and summarize information about the event handling, generating feedback text. Specifically, this includes: S31 generates a summarized context vector based on textual information and key elements extracted by the feature extraction model; S32 Based on pre-configured constraint optimization objectives, a conditional language model is used to generate an output sequence word by word to obtain feedback text. The constraint optimization objectives include the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and the preset standard template.

[0029] In step S31, the text information and extracted key elements are input into a pre-built and trained text induction model to obtain a context vector. This context vector is a high-dimensional, condensed semantic representation formed by the model after capturing the core meaning and task intent of the text information and key elements. Then, using the obtained context vector as a condition, an output sequence is generated word-by-word (or token-by-tag) to obtain the feedback text. During this process, the probability of the word output at each step is determined by a classifier such as softmax. The word with the highest probability (or sampled according to probability) is selected as the output of the current time step, ensuring that the generated feedback text is not only fluent but also highly semantically related to the input text information.

[0030] To improve the quality of generated feedback text, a constraint optimization objective was introduced, including fluency, completeness, conciseness, and standardization scores. The highest-scoring text (by multiplying these scores) was selected from candidate texts as the final feedback text, which was then filled into the feedback content input box on the event feedback page, completing the automatic filling of the event feedback. The fluency score is determined by the probability of each word in the feedback text; a higher probability indicates that the generated feedback text conforms to language habits, reads more naturally and smoothly, ensuring that the generated feedback text is not a rigid jumble of words but rather fluent and coherent language. The completeness score is determined by whether the feedback text contains preset key elements. Specifically, a Boolean indicator function is used to confirm whether the generated feedback text contains all necessary preset key elements; if all are contained, the function value is 1; otherwise, the function value is 0. The conciseness score is determined by the length of the feedback text, specifically by introducing a penalty term e. -λ|R| The measurement is performed where |R| is the length of the generated feedback text (e.g., word count), and λ is a moderating coefficient greater than 0. This value decreases exponentially as the length of the feedback text increases. The normativity score is determined by the semantic similarity between the feedback text and the preset standard template. Specifically, after converting the generated feedback text and the preset standard template into semantic vectors using models such as Sentence-BERT, the cosine of the angle between the vectors is calculated. The closer the cosine value is to 1, the more semantically similar the two are.

[0031] This embodiment is an improvement on the above embodiment. In this embodiment, after step S10, which involves obtaining the voice information input by the user based on the event feedback page and converting it into corresponding text information, the method further includes: S11 performs noise reduction processing on text information based on a preset noise reduction rule base; S12 uses a semantic analysis algorithm to identify semantic fragments in text information that are irrelevant to event feedback, and corrects semantically confusing expressions in text information.

[0032] In this embodiment, based on a preset noise reduction rule base and semantic analysis algorithm, the initial text information is denoised. The noise reduction rule base includes common interjections (such as "um," "ah," "that," etc.), filler words, meaningless filler words, and invalid text corresponding to background noise. The semantic analysis algorithm is used to identify semantic fragments in the text information that are not related to the event feedback (such as user chat content) and correct semantic confusion caused by speech transcription errors (such as saying "Road A is Route B" or "Identity Card" is "Identity Certificate," etc.). By deleting invalid text and irrelevant semantic fragments, clean and accurate target text information is obtained, and the target text information is transmitted to step S20 for element extraction. This can effectively improve the accuracy of key element extraction and reduce element recognition errors caused by text noise.

[0033] This embodiment is obtained by improving the above embodiments. In this embodiment, as follows: Figure 2 As shown, after step S40 fills the generated feedback text into the feedback content input box on the event feedback page, it also includes: S50 confirms whether the key elements extracted by the element extraction model are complete based on the pre-set list of necessary elements for event feedback. If all information is complete, the S60 should provide a submit button on the event feedback page for users to confirm the submission. If elements are incomplete, the S70 will display a missing element prompt on the event feedback page, allowing the user to supplement them, and then proceed to the step of confirming whether the key elements are complete after the user supplements them.

[0034] In this embodiment, a list of necessary elements for event feedback is preset, including ID card number, event location, event level, and processing result. These can be added or removed according to actual needs in the application. The key elements extracted in step S20 are compared with the list of necessary elements to determine if each necessary element is complete. If all elements are complete, a "complete elements" verification result is generated, and a submit button is provided on the event feedback page for user confirmation. If elements are incomplete, a "missing elements" verification result is generated, clearly indicating the missing element name, such as "missing ID card information" or "missing event level information," and the verification result and missing element information are displayed on the event feedback page, prompting the user to supplement them. After the user manually supplements the elements, the process proceeds to step S50 to confirm whether the supplemented key elements are complete.

[0035] In this embodiment, the system automatically determines whether the feedback elements are complete and promptly reminds the user to supplement the missing elements, avoiding incomplete feedback information due to user omissions. This provides complete and effective data support for subsequent event processing and reduces the workload of staff in verifying information.

[0036] Another embodiment of the present invention provides an event feedback filling system 100, such as... Figure 3 As shown, the system includes: a voice input module 110, used to acquire voice information input by the user based on the event feedback page; a voice-to-text module 120, connected to the voice input module 110, used to convert the voice information acquired by the voice input module 110 into corresponding text information; an element extraction module 130, connected to the voice-to-text module 120, used to extract key elements from the text information based on the element extraction model; a text summarization module 140, connected to the element extraction module 130, used to combine the text information and the key elements extracted by the element extraction model to perform information fusion and summarization of the event handling situation, and generate feedback text; and a text filling module 150, connected to the text summarization module 140, used to fill the generated feedback text into the feedback content input box of the event feedback page, completing the automatic filling of the event feedback.

[0037] In this embodiment, the event feedback page is a page in an application app that requires users to input feedback information about events. For example, it could be a page in an enterprise service app where employees report repairs, or a page in a campus service app where students report issues related to the cafeteria. The application app can also be a government service app, a lifestyle service app, etc. To lower the barrier to entry for users, a voice input entry is provided on the event feedback page, allowing users to record voice information about the event handling process using the built-in microphone of their terminal device. The recorded voice information is then sent to the voice-to-text module 120 to be converted into text information. The method for converting voice information into text information is not specifically limited here; any existing speech recognition algorithm can be used, such as a deep learning-based speech recognition model, to perform text transcription processing on the received voice information, converting it into the corresponding initial text information.

[0038] After obtaining the initial text information, the speech-to-text module 120 sends it to the feature extraction module 130. The module then inputs the pre-installed and trained feature extraction model to extract key features, performing semantic analysis and feature extraction on the text information. The feature extraction module 130 includes: an ID card number recognition unit, used to identify the user's ID card number based on pre-configured regular expressions; an event address recognition unit, used to identify place names in the text information and obtain the event address based on a pre-configured place name database; an event level recognition unit, used to identify keywords related to the event level in the text information and obtain the event level based on a pre-configured event level keyword database; and an event processing result recognition unit, used to extract keywords related to the event processing result in the text information and obtain the event processing result by combining the context of the keywords. Specifically: The process by which the ID card number recognition unit identifies a user's ID card number is as follows: A predefined regular expression for an 18-digit ID card number is used. Upon receiving text information, the unit searches for all consecutive strings within the text and checks whether each string can be completely matched by the defined regular expression. Finally, it returns all strings in the text that conform to the ID card number format, thus completing the matching of the user's ID card number.

[0039] The process by which the event address recognition unit identifies the address of an event is as follows: A place name database is pre-built, where each record is a structured standard address containing: province, city, district, street, and number. Upon receiving text information, a pre-built named entity recognition model is used to perform a preliminary scan of the text information, identifying and extracting all possible place name entity fragments (corresponding to the place names mentioned above), forming a list of place name entities. For example, from the text "I went to Zhongguancun Street, Haidian District, Beijing," a list of place name entities containing "Beijing," "Haidian District," and "Zhongguancun Street" may be identified. Next, a pre-configured similarity function is used to score the identified list of place name entities, finally outputting a standardized address vector containing the province, city, district, street, and number. It should be understood that the information contained in the mentioned standard address can be determined according to the actual situation; the above is just an example of one approach.

[0040] To improve the accuracy of event location extraction, similarity ( A , g Based on word overlap J ( A ∩name( g )) and hierarchical consistency L hier( A , g Together, we determine and finally extract the standard address with the highest similarity score for output: similarity A , g )= α • J ( A ∩name( g ))+ β • L hier( A , g ) in, A Represents a list of identified place-name entities; g This represents the standard address in the constructed place name database G; α and β Representing word overlap J ( A ∩name( g)) and hierarchical consistency L hier( A , g The weighting coefficients of ) and α + β =1, the specific value is determined based on the actual situation. Word overlap. J ( A ∩name( g This is used to measure the degree of textual overlap between place names identified in textual information and place names contained in standard addresses. J ( A ∩name( g ))=| A ∩name( g )| / A ∪name( g For example, the user-input text includes: Zhongguancun Street, Haidian District, Beijing; there is a record in the place name database: g = No. 27, Zhongguancun Street, Haidian District, Beijing; according to the standards of province, city, district, street, and number, the intersection is... A ∩name(g) = {Beijing, Beijing, Haidian, Zhongguancun Street}, Union A ∪name(g)={Beijing, Beijing, Haidian, Zhongguancun Street, No. 27}, J ( A ∩name( g =0.8. Hierarchical consistency L hier( A , g This is used to assess the degree of match / conflict between identified place names and candidate standard addresses in terms of administrative hierarchy. For example, if the user inputs text information including: Haidian District, Beijing; and there is a record in the place name database g = Chaoyang District, Beijing, according to the standard of province, city, district, street, and number, the two pieces of information are respectively (Beijing, Beijing, Haidian) and (Beijing, Beijing, Chaoyang), then... L hier( A , g =2 / 3.

[0041] The process by which the event level identification unit identifies event levels is as follows: A set of keywords is predefined for each event level, for example: emergency level event. K urgent = {immediate, right now, at the last minute}; matters of high importance. K high = {important, significant, serious, critical}; general level of event. K medium = {normal, common, regular}; minor level event K`low` is defined as {minor, minor issue, convenient, when available}, etc. After receiving the text message, the matching strength between it and each event level is calculated. Specifically, when calculating the matching strength with emergency level events, the emergency level events appearing in the text message are first counted. K The number of words in the urgent dictionary, then divided by the urgency level of the event. K The total number of words in the urgent dictionary, resulting in a value between 0 and 1, represents the intensity of the text's mention of that level. This process continues until the event level with the highest score is output. For example, if the text contains the word "urgent," the event level is urgent. K The urgent dictionary contains 4 words, so the matching strength S_level = 1 / 5 = 0.2.

[0042] To avoid situations where multiple event levels with the highest matching scores appear simultaneously, making it impossible to make a judgment, this embodiment further introduces context embedding vectors for disambiguation. Specifically: First, the entire text information is input into the BERT model or other similar models to obtain a high-dimensional vector representing its overall semantics. Then, the average semantic vector for each event level is obtained; for example, for emergency level events... K For urgent events, the average semantic vector for that level is obtained by averaging the BERT vectors of all words in the dictionary. Next, the cosine similarity between the high-dimensional vector of the text information and the average semantic vector of each event level is calculated. Finally, the closest event level is determined based on the calculated cosine similarity.

[0043] The process of identifying the event processing result by the event processing result identification unit is as follows: A corresponding keyword set is predefined for each processing result, for example: resolved status Kresolved = {resolved, completed}; processing status Kprocessing = {processing, accepted}; requiring coordination status Kpending = {requiring coordination, pending follow-up, further coordination needed}, etc. After receiving text information, the number of keywords appearing in the corresponding status keyword set is counted, and finally, the text information is assigned to the status result category with the most frequently appearing keywords. In other embodiments, lightweight classifiers such as softmax can also be used for classification.

[0044] After the element extraction module 130 extracts the key elements, it sends the mentioned key elements to the text summarization module 140 to perform information fusion and summarization on the event handling situation and generate feedback text. Specifically, the text summarization module 140 includes: a context vector summarization unit, used to generate a summarized context vector based on text information and key elements extracted by the element extraction model; and a feedback text generation unit, connected to the context vector summarization unit, used to generate an output sequence word by word using a conditional language model based on pre-configured constraint optimization objectives to obtain feedback text. The constraint optimization objectives include the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and a preset standard template.

[0045] In the context vector induction unit, textual information and extracted key elements are input into a pre-built and trained text induction model to obtain a context vector. This context vector is a high-dimensional, condensed semantic representation formed by the model after capturing the core meaning and task intent of the textual information and key elements. Then, the feedback text generation unit uses the obtained context vector as a condition to generate an output sequence word-by-word (or token-by-tag) to obtain the feedback text. During this process, the probability of the output word at each step is determined by a classifier such as softmax. The word with the highest probability (or sampled according to probability) is selected as the output of the current time step, ensuring that the generated feedback text is not only fluent but also highly semantically related to the input text information.

[0046] To improve the quality of generated feedback text, the feedback text generation process incorporates constrained optimization objectives including fluency, completeness, conciseness, and standardization scores. The text with the highest score (the product of these scores) is selected from candidate texts and used as the final feedback text. This generated feedback text is then filled into the feedback content input box on the event feedback page, completing the automatic filling of the event feedback. The fluency score is determined by the probability of each word in the feedback text. A higher probability indicates that the generated feedback text conforms to language habits, reads more naturally and smoothly, ensuring that the generated feedback text is not a rigid jumble of words but rather fluent and coherent language. The completeness score is determined by whether the feedback text contains preset key elements. Specifically, a Boolean indicator function checks whether the generated feedback text contains all necessary preset key elements. If all are contained, the function returns a value of 1; otherwise, the function returns a value of 0. The conciseness score is determined by the length of the feedback text, specifically by introducing a penalty term e. -λ|R|The measurement is performed where |R| is the length of the generated feedback text (e.g., word count), and λ is a moderating coefficient greater than 0. This value decreases exponentially as the length of the feedback text increases. The normativity score is determined by the semantic similarity between the feedback text and the preset standard template. Specifically, after converting the generated feedback text and the preset standard template into semantic vectors using models such as Sentence-BERT, the cosine of the angle between the vectors is calculated. The closer the cosine value is to 1, the more semantically similar the two are.

[0047] This embodiment is obtained by improving the above embodiment. In this embodiment, the event feedback filling system 100 further includes a text denoising module connected to the speech-to-text module 120 and the element extraction module 130 respectively. The text denoising module is used to perform denoising processing on the text information based on a preset denoising rule base, and to identify semantic segments in the text information that are not related to the event feedback based on a semantic analysis algorithm, and to correct semantically confused expressions in the text information.

[0048] In this embodiment, the initial text information is denoised based on a preset noise reduction rule base and semantic analysis algorithm. The noise reduction rule base includes common interjections (such as "um," "ah," "that," etc.), filler words, meaningless filler words, and invalid text corresponding to background noise. The semantic analysis algorithm is used to identify semantic fragments in the text information that are not related to the event feedback (such as user chat content) and correct semantic confusion caused by speech transcription errors (such as saying "Road A is Route B" or "Identity Card" is "Identity Certificate," etc.). By deleting invalid text and irrelevant semantic fragments, clean and accurate target text information is obtained, and the target text information is transmitted to the element extraction module 130 for element extraction. This can effectively improve the accuracy of key element extraction and reduce element recognition errors caused by text noise.

[0049] This embodiment is an improvement on the above embodiments. In this embodiment, the event feedback filling system 100 further includes: an element verification module, connected to the element extraction module 130, used to confirm whether the key elements extracted by the element extraction model are complete based on a preset list of necessary elements for event feedback; a submission module, connected to the element verification module, used to provide a submission button on the event feedback page for the user to confirm submission if the element verification module confirms that the elements are complete; and a manual supplementation module, connected to the element verification module, used to display a missing element prompt on the event feedback page for the user to supplement if the element verification module confirms that the elements are incomplete, and to send the supplemented key elements to the element verification module after the user supplements them.

[0050] In this embodiment, a list of necessary elements for event feedback is preset, including ID card number, event location, event level, and processing result. These can be added to or removed from the list based on actual needs. The element verification module compares the key elements extracted by the element extraction module 130 with the list of necessary elements to determine if each necessary element is complete. If all elements are complete, a verification result of "Element Complete" is generated and sent to the submission module; if elements are incomplete, a verification result of "Element Missing" is generated, clearly indicating the missing element name, such as "Missing ID Card Information" or "Missing Event Level Information," and the verification result and missing element information are sent to the manual supplementation module.

[0051] After receiving the "all elements are complete" verification result from the element verification module, the submission module automatically retrieves the feedback text and all key elements of the event handling status already filled in on the event feedback page, generating complete event feedback information. At the same time, it provides the user with a "Submit" button. After the user clicks this button, the submission module transmits the complete event feedback information to the APP backend server, completing the event feedback submission operation. After successful submission, a "Submission Successful" prompt pops up on the APP page to inform the user of the feedback result.

[0052] After receiving the "Missing Element" verification result and missing element information from the element verification module, the manual supplementation module displays a prompt window on the event feedback page, reminding the user to supplement the missing feedback elements. Simultaneously, it provides the user with a text input box or drop-down selection box (e.g., "Urgent," "Important," "General," "Minor" drop-down options for the event level) to facilitate quick information supplementation. Once the user has supplemented the information, clicking the "Confirm Supplementation" button will send the supplemented element data to the element verification module for secondary verification until all elements are complete.

[0053] In another embodiment of the present invention, such as Figure 4 As shown, a terminal device 200 includes a processor 210 and a memory 220. The memory 220 is used to store a computer program 221. The processor 210 is used to execute the computer program 221 stored in the memory 220 to implement the method in the above-mentioned event feedback filling method embodiment.

[0054] Terminal device 200 may include, but is not limited to, processor 210 and memory 220. Those skilled in the art will understand that... Figure 4This is merely an example of terminal device 200 and does not constitute a limitation on terminal device 200. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, terminal device 200 may also include input / output interfaces, display devices, network access devices, communication buses, communication interfaces, etc. The communication interfaces and communication buses may also include input / output interfaces, wherein the processor 210, memory 220, input / output interfaces, and communication interfaces communicate with each other through the communication bus. The memory 220 stores a computer program 221, and the processor 210 executes the computer program 221 stored in the memory 220 to implement the methods in the corresponding method embodiments described above.

[0055] The processor 210 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0056] The memory 220 can be an internal storage unit of the terminal device 200, such as a hard disk or RAM. The memory can also be an external storage device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or FlashCard. Furthermore, the memory 220 can include both internal and external storage units of the terminal device 200. The memory 220 is used to store the computer program 221 and other programs and data required by the terminal device 200. The memory can also be used to temporarily store data that has been output or will be output.

[0057] A communication bus is a circuit that connects the described elements and enables transmission between these elements. For example, processor 210 receives commands from other elements via the communication bus, decrypts the received commands, and performs calculations or data processing based on the decrypted commands. Memory 220 may include program modules, for example, a kernel, middleware, an application programming interface (API), and applications. The program module may consist of software, firmware, or hardware, or at least two of these. Input / output interfaces forward commands or data input by the user through input / output interfaces (for example, sensors, keyboards, touchscreens). Communication interfaces connect terminal device 200 to other network devices, user equipment, and networks. For example, the communication interface may connect to a network via wired or wireless means to connect to other external network devices or user equipment. Wireless communication may include at least one of the following: Wi-Fi, Bluetooth (BT), Near Field Communication (NFC), Global Positioning System (GPS), and cellular communication, etc. Wired communication may include at least one of the following: Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), Asynchronous Transfer Standard (RS-232), etc. The network may be a telecommunications network or a communication network. The communication network may be a computer network, the Internet, the Internet of Things (IoT), or a telephone network. Terminal device 200 can connect to the network through a communication interface, and the protocol used by terminal device 200 to communicate with other network devices may be supported by at least one of the following: application programming interface (API), middleware, kernel, and communication interface.

[0058] In another embodiment, the present invention also provides a storage medium storing at least one instruction, which is loaded and executed by a processor to perform the operations described in the corresponding embodiments above. For example, the storage medium may be a read-only memory (ROM), a random access memory (RAM), a read-only optical disc (CD-ROM), a magnetic tape, a floppy disk, or an optical data storage device, etc.

[0059] These can be implemented using computer-executable program code, thus allowing them to be stored in a storage device for execution by a computing device, or fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, this invention is not limited to any particular hardware and software combination.

[0060] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0061] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0062] In the embodiments provided in this application, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative; the division of modules or units is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interface; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0063] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0064] Furthermore, the functional units in the various embodiments of this application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0065] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the present invention can implement all or part of the processes in the methods of the above embodiments by sending instructions to related hardware via a computer program 221. The computer program 221 can be stored in a storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program 221 can be in the form of source code, object code, executable file, or some intermediate form. The storage medium can include: any entity or device capable of carrying the computer program 221, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0066] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An event feedback filling method, characterized in that, include: Obtain the voice information input by the user based on the event feedback page and convert it into corresponding text information; Key elements are extracted from the text information based on the element extraction model. By combining the text information and the key elements extracted by the element extraction model, information fusion and summarization of the event handling situation are performed to generate feedback text. The generated feedback text is then filled into the feedback content input box on the event feedback page to complete the automatic filling of the event feedback.

2. The event feedback filling method of claim 1, wherein, The extraction of features from the text information based on the feature extraction model includes: The user's ID card number is identified based on pre-configured regular expression matching. Identify place names in the text information and obtain the address where the event occurred based on a pre-configured place name database; Identify keywords related to event levels in the text information and obtain the event level based on a pre-configured event level keyword library; Extract keywords related to the event processing result from the text information, and obtain the event processing result by combining the context of the keywords.

3. The event feedback filling method as described in claim 1, characterized in that, Combining the text information and the key elements extracted by the element extraction model, information fusion and summarization are performed on the event handling situation to generate feedback text, including: A summarized context vector is generated based on the text information and the key elements extracted by the element extraction model. Based on pre-configured constraint optimization objectives, a conditional language model is used to generate an output sequence word by word to obtain the feedback text. The constraint optimization objectives include the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and a preset standard template.

4. The event feedback filling method as described in any one of claims 1-3, characterized in that, After filling the generated feedback text into the feedback content input box on the event feedback page, the process also includes: Based on the pre-set list of necessary elements for event feedback, confirm whether the key elements extracted by the element extraction model are complete; To ensure all information is complete, a submit button should be provided on the event feedback page for users to confirm their submission. If an element is incomplete, a missing element prompt will be displayed on the event feedback page for the user to supplement. After the user supplements the missing element, the user will be prompted to re-enter the step to confirm whether the key element is complete.

5. An event feedback filling system, characterized in that, include: The voice input module is used to acquire voice information input by the user based on the event feedback page; A speech-to-text module, connected to the speech input module, is used to convert the speech information acquired by the speech input module into corresponding text information; The element extraction module, connected to the speech-to-text module, is used to extract key elements from the text information based on the element extraction model. The text summarization module, connected to the element extraction module, is used to combine the text information and the key elements extracted by the element extraction model to perform information fusion and summarization on the event handling situation and generate feedback text. The text filling module, connected to the text summarization module, is used to fill the generated feedback text into the feedback content input box of the event feedback page, thereby completing the automatic filling of the event feedback.

6. The event feedback filling system as described in claim 5, characterized in that, The feature extraction module includes: The ID card number recognition unit is used to identify the user's ID card number based on a pre-configured regular expression. An event address identification unit is used to identify place names in the text information and obtain the event location based on a pre-configured place name database. An event level identification unit is used to identify keywords related to event levels in the text information and obtain the event level based on a pre-configured event level keyword library; The event processing result recognition unit is used to extract keywords related to the event processing result from the text information and obtain the event processing result by combining the context of the keywords.

7. The event feedback filling system as described in claim 5, characterized in that, The text summarization module includes: The context vector induction unit is used to generate an inductive context vector based on the text information and the key elements extracted by the element extraction model. The feedback text generation unit, connected to the context vector induction unit, is used to generate an output sequence word by word using a conditional language model based on a pre-configured constraint optimization objective, thereby obtaining the feedback text. The constraint optimization objective includes the fluency score, completeness score, conciseness score, and standardization score of the feedback text. The fluency score is determined by the probability of each word in the feedback text, the completeness score is determined by whether the feedback text contains preset key elements, the conciseness score is determined by the length of the feedback text, and the standardization score is determined by the semantic similarity between the feedback text and a preset standard template.

8. The event feedback filling system as described in any one of claims 5-7, characterized in that, The event feedback population system also includes: The element verification module, connected to the element extraction module, is used to confirm whether the key elements extracted by the element extraction model are complete based on a list of necessary elements fed back by a preset event. The submission module is connected to the element verification module. If the element verification module confirms that all elements are complete, it provides a submission button on the event feedback page for the user to confirm the submission. The manual supplementation module is connected to the element verification module. If the element verification module confirms that the elements are incomplete, it displays a missing element prompt on the event feedback page for the user to supplement, and sends the supplemented key elements to the element verification module after the user supplements.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor runs the computer program, it implements the steps of the event feedback filling method as described in any one of claims 1-4.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the event feedback filling method as described in any one of claims 1-4.

Images