Real-time semantic analysis and dynamic reply template matching device for advertiser consultation

Abstract

The invention provides a real-time semantic analysis and dynamic reply template matching device for advertisement customer consultation, and relates to the technical field of reply template matching. A scene database, a scene event database and a scene event function database of a reply information base are obtained; acquiring user input processing data, further acquiring a matching scene database, a matching event database and a matching function database, and further acquiring a user matching database and matching output data; obtaining a preset template database, further obtaining a scene template library, a template event library and a template function library, and obtaining a matched scene template library, a matched event template library, a matched function template library and a matched function template according to the matched output data; obtaining an initial matching function module and a newly-added matching function module, obtaining a template similarity difference value, judging intention recognition error information, and performing parameter adjustment. By means of the method and device, high-precision template matching and replying of consultation information are achieved.

Description

Technical Field

[0001] The present invention proposes a real-time semantic analysis and dynamic reply template matching device for advertiser consultation, which relates to the technical field of reply template matching, and specifically to the technical field of real-time semantic analysis and dynamic reply template matching. Background Art

[0002] Traditional semantic analysis responses are usually based on the comprehensive processing of big data, lack of specificity, and can easily lead to misunderstandings of the intent of customer consultation information, which in turn leads to the output of erroneous information. It is difficult to discover and adjust in a timely manner, and it is often difficult to detect in time when the user's intention changes. As a result, traditional methods have obvious deficiencies in template matching accuracy, dynamic adjustment capabilities, intent recognition error detection, resource utilization efficiency and quantitative analysis, and are unable to meet the needs of modern intelligent systems for precise matching and efficient optimization. Summary of the Invention

[0003] The present invention provides a real-time semantic analysis and dynamic response template matching device for advertiser consultation to solve the above problems: The present invention proposes a device for real-time semantic analysis and dynamic response template matching for advertiser consultation, the device comprising: The library management module is used to obtain the scene database, scene event database and scene event function database of the reply information library; An input data matching module is used to obtain user input processing data, and then obtain a matching scenario database, a matching event database, and a matching function database, and then obtain a user matching database and matching output data; The output template matching module is used to obtain the preset template database, and then obtain the scene template library, template event library and template function library, and obtain the matching scene template library, matching event template library, matching function template library and matching function template according to the matching output data; The template correction module is used to obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, judge the intent recognition error information, and adjust the parameters.

[0004] Furthermore, the warehouse management module includes: The scenario classification module is used to construct a response information database based on big data information, classify the data scenarios of the response information database, and obtain a scenario database; An event classification module is used to classify data events in the scene database and obtain a scene event database of the scene database; A function classification module is used to perform data function classification on the scene event database to obtain a scene event function database of the scene database; Furthermore, the input data matching module includes: A pre-training module is used to obtain user input information, perform pre-training processing on the user input information, and obtain user input processing data; A scene matching module, configured to match the user input processing data with a scene database to obtain a matching scene database; An event matching module, configured to match the user input processing data with a scene event database of a matching scene database to obtain a matching event database of the matching scene database; A function matching module, configured to match the user input processing data with a matching scene database, a matching event database, and a matching function database to obtain a matching scene database, a matching event database, and a matching function database; The matching library construction module is used to generate a user matching database based on the matching scenario database, the matching event database and the matching function database.

[0005] The model output module is used to train the intent classification model through the user matching database and obtain matching output data.

[0006] Furthermore, the model output module includes: A model training module is used to train the intent classification model through the scene database, the scene event database, and the scene event function database; A data output module, configured to input the user matching database into an intent classification model to obtain matching output data; Furthermore, the output template matching module includes: A scene template acquisition module is used to obtain a preset template database, classify the preset template database into template scenes, and obtain a scene template library; An event template acquisition module, configured to classify the scene template library into template events and obtain a template event library of the scene template library; A function template acquisition module, configured to classify the template function of the template event library to obtain a template function library of a scene template library, a template event library, and a template function library; Furthermore, the output template matching module also includes: A scene template matching module, configured to match the matching output data with a scene template library to obtain a matching scene template library; An event template acquisition module, configured to match the matching output data with the template event library of the scene template library through the matching scene template library, and obtain a matching event template library of the matching scene template library; The function template acquisition module is used to match the matching output data with the template function library of the scene template library, the template event library and the template function library through the matching event template library to obtain the matching function template of the matching scene template library, the matching event template library and the matching function template library.

[0007] Furthermore, the template correction module includes: The initial filling module is used to retrieve the initial matching function template from the preset template database by matching the output data; Filling variables of the initial matching function template with the matching output data to obtain a target output template; A new matching module is added, which is used to obtain new user input information and obtain new matching output data through the new user input information; Retrieve a new matching function template from the preset template database according to the new matching output data; The intent recognition analysis module is used to calculate template similarity, determine the intent recognition status based on the template similarity, and obtain intent recognition information.

[0008] Furthermore, the intention recognition and analysis module also includes: A similarity calculation module is used to obtain the template similarity between the initial matching function template and the newly added matching function template; A difference acquisition module is used to obtain a preset similarity threshold, calculate the difference between the preset similarity threshold and the template similarity, and obtain the template similarity difference; A difference error recognition module is used to compare the template similarity difference with a preset similarity difference threshold to obtain a difference comparison result; Determine whether there is an intention recognition error based on the difference comparison result, and obtain error recognition information; When there is erroneous recognition information, the matching function template is reacquired until there is no intention recognition error.

[0009] Furthermore, the difference error identification module includes: When the template similarity difference is greater than the preset similarity difference threshold, it is determined that there is an intent recognition error; When the template similarity difference is less than or equal to the preset similarity difference threshold, it is determined that there is no intention recognition error; When the template similarity difference is greater than a preset similarity difference threshold, obtaining a ratio of the template similarity difference to the preset similarity difference threshold to obtain a matching deviation coefficient; The multiple of the adjustment parameter of the user input processing data is determined based on the matching deviation coefficient.

[0010] Furthermore, the matching method includes: Obtaining a scenario database, a scenario event database, and a scenario event function database of a response information database; Obtain user input processing data, and then obtain matching scenario database, matching event database and matching function database, and then obtain user matching database and matching output data; Obtain a preset template database, and then obtain a scene template library, a template event library, and a template function library, and obtain a matching scene template library, a matching event template library, a matching function template library, and a matching function template according to the matching output data; Obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, determine the intent recognition error information, and adjust the parameters.

[0011] Beneficial effects of the present invention: The device classifies and matches user input information consulted by advertisers to a database, obtains a matching database, trains an intent classification model through the matching database, outputs matching output data, matches templates of scenes, events, and functions through the matching output data, calculates template similarity, determines whether the template output is accurate or whether the user has changed his intent, and corrects the template; The library management module categorizes the response information library by scenario, event, and function, obtaining classified response information for each scenario, event, and function. This enables coarse-to-fine classification management of the response information by scenario, event, and function, improving the flexibility of coarse-grained data processing.

[0012] The input data matching module obtains accurate input data, matches the input data with scenarios, events, and functions, and then inputs the matched model data. By performing multi-dimensional data matching on the input data with scenarios, events, and functions, matching data that closely matches the user input data is obtained and input into the model, achieving accurate extraction of the input data. This improves the accuracy of the model output data from the source, and further enhances the relevance and fit of the output data. The output template matching module is used to match the output data's scenes, events, and functions. By matching the output data's scenes, events, and functions, multi-dimensional template matching can be performed on the output data, ensuring that each part of the obtained template corresponds to the output data, thereby enhancing the accuracy of template matching. The template correction module can identify mismatched templates or the templates that the user intended to change, and then correct and adjust the erroneous templates. Similarity calculation and correction further identify and correct erroneous templates, achieving automated optimized template matching and improving template correction efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1Schematic diagram of a real-time semantic analysis and dynamic response template matching device for advertising consultation; Figure 2 Schematic diagram of the real-time semantic analysis and dynamic response template matching method for advertising consultation. DETAILED DESCRIPTION

[0014] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0015] In one embodiment of the present invention, a device for real-time semantic analysis and dynamic response template matching for advertiser inquiries is provided, the device comprising: The library management module is used to obtain the scene database, scene event database and scene event function database of the reply information library; An input data matching module is used to obtain user input processing data, and then obtain a matching scenario database, a matching event database, and a matching function database, and then obtain a user matching database and matching output data; The output template matching module is used to obtain the preset template database, and then obtain the scene template library, template event library and template function library, and obtain the matching scene template library, matching event template library, matching function template library and matching function template according to the matching output data; The template correction module is used to obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, judge the intention recognition error information, and adjust the parameters, such as Figure 1 shown.

[0016] The working principle of the above technical solution is: the library management module is used to obtain the scene database, scene event database and scene event function database of the reply information library; the reply information library is classified by scenes, events and functions through the library management module to obtain classified reply information of scenes, events and functions.

[0017] The input data matching module is used to obtain user input processing data, and then obtain the matching scene database, matching event database and matching function database, and then obtain the user matching database and matching output data; the input data matching module obtains accurate input data, matches the input data with scenes, events and functions, and inputs the matched model data; The output template matching module is used to obtain the preset template database, and then obtain the scene template library, template event library and template function library, and obtain the matching scene template library, matching event template library, matching function template library and matching function template according to the matching output data; the output template matching module is used to perform template matching of scenes, events and functions of the matching output data; The template correction module is used to obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, determine the intent recognition error information, and adjust the parameters. The template correction module can obtain the template with mismatch or the template corresponding to the user's intention to change, and correct and adjust the incorrect template.

[0018] The technical effects of the above technical solution are as follows: the device classifies and matches the user input information of the advertiser's consultation in the database, obtains a matching database, trains the intent classification model through the matching database, outputs matching output data, matches the matching output data with templates of scenes, events and functions, calculates the template similarity, determines whether the template output is accurate or whether the user has changed the intent, and corrects the template; The library management module categorizes the response information library by scenario, event, and function, obtaining classified response information for each scenario, event, and function. This enables coarse-to-fine classification management of the response information by scenario, event, and function, improving the flexibility of coarse-grained data processing.

[0019] The input data matching module obtains accurate input data, matches the input data with scenarios, events, and functions, and then inputs the matched model data. By performing multi-dimensional data matching on the input data with scenarios, events, and functions, matching data that closely matches the user input data is obtained and input into the model, achieving accurate extraction of the input data. This improves the accuracy of the model output data from the source, and further enhances the relevance and fit of the output data. The output template matching module is used to match the output data's scenes, events, and functions. By matching the output data's scenes, events, and functions, multi-dimensional template matching can be performed on the output data, ensuring that each part of the obtained template corresponds to the output data, thereby enhancing the accuracy of template matching. The template correction module can identify mismatched templates or the templates that the user intended to change, and then correct and adjust the erroneous templates. Similarity calculation and correction further identify and correct erroneous templates, achieving automated optimized template matching and improving template correction efficiency and accuracy.

[0020] In one embodiment of the present invention, the warehouse management module includes: The scenario classification module is used to construct a response information database based on big data information, classify the data scenarios of the response information database, and obtain a scenario database; An event classification module is used to classify data events in the scene database and obtain a scene event database of the scene database; The function classification module is used to perform data function classification on the scene event database to obtain the scene event function database of the scene database.

[0021] The working principle of the above technical solution is as follows: the scenario classification module is used to construct a response information database through big data information, classify the data scenarios of the response information database, and obtain a scenario database; The scenario database includes advertising recommendation scenarios, transaction scenarios, and transportation scenarios, etc. An event classification module is used to classify data events in the scene database and obtain a scene event database of the scene database; The scenario event database includes click analysis events, conversion rate optimization events, and budget allocation events in the advertising recommendation scenario; order payment time, refund application time, and product consultation events in the transaction scenario; and shipping delay events, logistics query events, and package loss events in the transportation scenario. The function classification module is used to perform data function classification on the scene event database to obtain the scene event function database of the scene database.

[0022] The scenario event function database includes data such as logistics query, delay notification and complaint handling of transportation event functions. The transaction event function includes payment confirmation, refund processing, order inquiry and other data; the advertising event function includes click-through rate analysis, budget optimization, creative recommendation and other data.

[0023] The technical effect of the above technical solution is: by classifying the response information library according to the scene, an information library of multiple scenes can be obtained, and targeted response analysis of each scene can be achieved, thereby improving the accuracy and efficiency of scene recognition.

[0024] By classifying the response information library according to the events of the scene, an information library of events of multiple scenes can be obtained, and targeted response analysis of each event of each scene can be achieved, thereby improving the accuracy and efficiency of identifying the events of the scene.

[0025] By classifying the response information library according to the functions of the events in the scene, an information library of the functions of the events in multiple scenes can be obtained, and targeted response analysis of the functions of each event in each scene can be achieved, thereby improving the accuracy and efficiency of identifying the functions of the events in the scene.

[0026] In one embodiment of the present invention, the input data matching module includes: A pre-training module is used to obtain user input information, perform pre-training processing on the user input information, and obtain user input processing data; A scene matching module, configured to match the user input processing data with a scene database to obtain a matching scene database; An event matching module, configured to match the user input processing data with a scene event database of a matching scene database to obtain a matching event database of the matching scene database; A function matching module, configured to match the user input processing data with a matching scene database, a matching event database, and a matching function database to obtain a matching scene database, a matching event database, and a matching function database; The matching library construction module is used to generate a user matching database based on the matching scenario database, the matching event database and the matching function database.

[0027] The model output module is used to train the intent classification model through the user matching database and obtain matching output data.

[0028] The working principle of the above technical solution is as follows: the pre-training module is used to obtain user input information, perform pre-training processing on the user input information, and obtain user input processing data; A scene matching module is used to match the user input processing data with a scene database to obtain a matching scene database; the matching scene database is a scene database suitable for the user input information; An event matching module, configured to match the user input processing data with a scene event database of a matching scene database to obtain a matching event database of the matching scene database; the matching event database is an event database applicable to the user input information; a function matching module, configured to match the user input processing data with a matching scene database, a matching event database, and a scene event function database to obtain a matching function database that matches the matching scene database and the matching event database; the matching function database is a function database applicable to the user input information; The matching database construction module is used to generate a user matching database based on the matching scene database, the matching event database and the matching function database. The user matching database includes scene data, event database and function database applicable to user input information.

[0029] The model output module is used to train the intent classification model through the user matching database and obtain matching output data.

[0030] The pre-training includes: The user input information is pre-trained and processed, and the pre-training processing includes operations such as data cleaning, word segmentation, and vectorization to generate user input processing data.

[0031] Specifically include: Remove noisy data (such as irrelevant characters, stop words, etc.).

[0032] Divide text information into meaningful vocabulary units.

[0033] Convert text or image information into numerical vectors (such as through word embedding models Word2Vec, BERT, or image feature extraction models ResNet).

[0034] The technical effects of the above technical solution are: through pre-training processing, the recognition sensitivity of user input information can be enhanced, and more efficient data response matching of user input information can be achieved, ensuring that user input information is easier to recognize; By matching the scenario database with the user input processing data, a scenario database with response information that closely matches the user input processing data can be obtained, which enhances the pertinence of the scenario response of the user input processing data, reduces the probability of responding to the wrong scenario, and improves the accuracy of the scenario response of the user input processing data; By matching the scene event database of the scene database with the user input processing data, a scene event database of the scene database that closely matches the reply information of the user input processing data can be obtained, thereby enhancing the pertinence of the reply to the event of the scene of the user input processing data, reducing the probability of replying to the event of the wrong scene, and improving the accuracy of the reply to the scene event of the user input processing data; By matching the scene event database, the scene event function database, and the scene database with the user input processing data, a scene event function database of the scene database that closely matches the reply information of the user input processing data can be obtained, thereby enhancing the pertinence of the reply function of the event of the scene of the user input processing data, reducing the probability of replying the function of the event of the wrong scene, and improving the accuracy of the reply function of the scene event of the user input processing data; The matching output data output by the intent classification model trained by the user matching database is highly targeted and consistent with the user matching database, so that the output data and the input data are highly correlated in terms of scenarios, events and functions.

[0035] In one embodiment of the present invention, the model output module includes: Model training module, used to train the intent classification model using data such as the scene database, scene event database, and scene event function database; A data output module, configured to input the user matching database and other data into the intent classification model to obtain matching output data; The working principle of the above technical solution is as follows: the model training module is used to train the intent classification model through the scene database, the scene event database and the scene event function database; the intent classification model trained by the scene database, the scene event database and the scene event function database adopts precise data categories and data hierarchies; The data output module is used to input the user matching database into the intent classification model to obtain matching output data; perform data analysis on the user matching database through the intent classification model, and output corresponding data for accurate data matching analysis.

[0036] Some of the code for the above content includes: # Load data scenes_df = pd.read_csv('scenes.csv') # scene database events_df = pd.read_csv('events.csv') # Scene event database functions_df = pd.read_csv('functions.csv') # Scene event function database # Merge data training_data = merge_databases(scenes_df, events_df, functions_df) # Feature Engineering: Converting Text Data into TF-IDF Vectors vectorizer = TfidfVectorizer(max_features=5000) X = vectorizer.fit_transform(training_data['description']) # Assume that the description field contains text description y = training_data['category'] # Assume that the category field is the target category (scene, event, function) # Divide the training set and test set X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) # Training intent classification model (taking random forest as an example) model = RandomForestClassifier(n_estimators=100, random_state=42) model.fit(X_train, y_train) # Evaluate the model y_pred = model.predict(X_test)print(f"Model accuracy: {accuracy_score(y_test, y_pred)}").

[0037] The technical effects of the above technical solution are: the intent classification model obtained by training the scene database, scene event database and scene event function database adopts accurate data categories and data hierarchies, and realizes the use of less data to train the model and output the most accurate output data; The user matching database is analyzed through the intent classification model, and the corresponding data is output, which realizes accurate matching analysis of the data and closely fits the corresponding scenarios, events and functions, so that the model output data is more consistent with the scenarios, events and functions of the user input information.

[0038] In one embodiment of the present invention, the output template matching module includes: A scene template acquisition module is used to obtain a preset template database, classify the preset template database into template scenes, and obtain a scene template library; An event template acquisition module, configured to classify the scene template library into template events and obtain a template event library of the scene template library; The function template acquisition module is used to classify the template functions of the template event library and obtain the template function library of the template event library of the scene template library.

[0039] The working principle of the above technical solution is: obtain the preset template database, and then obtain the scene template library, template event library and template function library, and obtain the matching scene template library, matching event template library, matching function template library and matching function template according to the matching output data; Specifically include: A scene template acquisition module is used to obtain a preset template database, classify the preset template database into template scenes, and obtain a scene template library; the scene template library includes scene template libraries corresponding to multiple scenes in the scene database; An event template acquisition module is used to classify the scene template library into template events to obtain a template event library of the scene template library; the template event library includes event template libraries corresponding to multiple events of multiple scene event databases of the scene database; The function template acquisition module is used to classify the template function of the template event library to obtain the template function library of the scene template library, the template event library, and the template function library; the template function library includes function template libraries corresponding to multiple functions of the scene database, the scene event database, and the multiple scene event function database.

[0040] The technical effects of the above technical solution are as follows: the scene template library includes scene template libraries corresponding to multiple scenes in the scene database; each preset scene is classified into scene template types to obtain template data for each scene, thereby improving the accuracy and independence of the scene templates, enhancing the scheduling efficiency of the scene templates, reducing scheduling resources, and enhancing the pertinence and accuracy of the scene template supply; The template event library includes an event template library corresponding to multiple events in a scene event database of a scene database; the scene event template type is classified for each preset scene event to obtain template data for each scene event, thereby improving the accuracy and independence of the scene event template, enhancing the scheduling efficiency of the scene event template, reducing scheduling resources, and enhancing the pertinence and accuracy of the scene event template supply; The template function library includes a function template library corresponding to multiple functions of multiple scene event function databases in the scene event database. The template event library includes an event template library corresponding to multiple events in the scene event database. The functions of each preset scene event are classified into scene event function template types to obtain template data for each scene event function, thereby improving the accuracy and independence of the scene event function templates, enhancing the scheduling efficiency of the scene event function templates, reducing scheduling resources, and enhancing the pertinence and accuracy of the supply of scene event function templates.

[0041] In one embodiment of the present invention, the output template matching module further includes: A scene template matching module, configured to match the matching output data with a scene template library to obtain a matching scene template library; An event template acquisition module, configured to match the matching output data with the template event library of the scene template library through the matching scene template library, and obtain a matching event template library of the matching scene template library; The function template acquisition module is used to match the matching output data with the template function library of the scene template library, the template event library and the template function library through the matching event template library to obtain the matching function template of the matching scene template library, the matching event template library and the matching function template library.

[0042] The working principle of the above technical solution is as follows: the scene template matching module is used to match the matching output data with the scene template library to obtain a matching scene template library; the matching scene template library includes the scene template library that matches the matching output data; An event template acquisition module is configured to match the matching output data with the template event library of the scene template library through the matching scene template library to obtain a matching event template library of the matching scene template library; the matching event template library includes the template event library in the matching scene template library that matches the matching output data; The function template acquisition module is configured to match the matching output data with the template function library of the scenario template library, the template event library, and the template function library using the matching event template library to obtain matching function templates of the matching scenario template library, the matching event template library, and the matching function template library. The matching function templates include the matching function templates of the matching event template library in the matching scenario template library that best match the matching output data.

[0043] The technical effects of the above technical solution are: obtaining scene template library data that matches the matching output data, and on the basis of associating the user input data with the matching output data, closely associating the matching output data with the scene template library, thereby enabling the user input data to be associated with the scene template library, thereby realizing a scene template library that is closely associated with the user input data and the matching output data, greatly enhancing the relevance of the selection of the scene template for the user input data, and improving the accuracy, relevance and intelligence of the template selection; Acquire event template library data of the scene that matches the matching output data, and on the basis of associating the user input data with the matching output data, closely associate the matching output data with the event template library of the scene, thereby enabling the user input data to be associated with the event template library of the scene, thereby realizing an event template library for the scene that is closely associated with the user input data and the matching output data, greatly enhancing the relevance of the selection of event templates for the scene of the user input data, and improving the accuracy, relevance and intelligence of template selection; Acquire the function template library data of the event of the scene that matches the matching output data, and on the basis of associating the user input data with the matching output data, tightly associate the matching output data with the function template library of the scene's event, thereby enabling the user input data to be associated with the function template library of the scene's event, thereby realizing an event template library for the scene that is tightly associated with the user input data and the matching output data, greatly enhancing the relevance of the selection of the function template of the event of the scene of the user input data, and improving the accuracy, relevance and intelligence of the template selection.

[0044] In one embodiment of the present invention, the template correction module includes: The initial filling module is used to retrieve the initial matching function template from the preset template database by matching the output data; Filling variables of the initial matching function template with the matching output data to obtain a target output template; A new matching module is added, which is used to obtain new user input information and obtain new matching output data through the new user input information; Retrieve a new matching function template from the preset template database according to the new matching output data; The intent recognition analysis module is used to calculate template similarity, determine the intent recognition status based on the template similarity, and obtain intent recognition information.

[0045] The working principle of the above technical solution is as follows: the initial filling module is used to retrieve the initial matching function template from the preset template database by matching the output data; the initial matching function template is the matching function template obtained before the newly added matching function template; Filling the variables of the initial matching function template with the matching output data to obtain a target output template; the target output template is a targeted complete template obtained from the matching output data; A new matching module is added, which is used to obtain new user input information and obtain new matching output data through the new user input information; the new matching output data is the output data re-matched according to the user feedback information; The newly added matching function template is retrieved from the preset template database based on the newly added matching output data; the newly added matching function template is used to calibrate and identify errors of the last matched template; and the corresponding template can also be changed when the user's intention changes.

[0046] The intent recognition analysis module is used to calculate template similarity, determine the intent recognition status based on the template similarity, and obtain intent recognition information.

[0047] The technical effect of the above technical solution is that: the initial matching function template is the matching function template obtained before the newly added matching function template; the initial matching function template retrieved and matched with the matching output data can be used as the basis for template comparison to determine whether the template selection is different from the user's intention; The target output template is a targeted complete template obtained from the matching output data; obtaining a targeted complete template ensures the integrity and replyability of the matching template, and improves the accuracy and professionalism of the reply; The newly added matching output data is the output data re-matched according to the user feedback information; through the newly added matching output data, the user's satisfaction with the previous template can be obtained, and then the deviation of the previous template output can be obtained.

[0048] New matching function templates are used to calibrate and identify errors in the previous matching template; they can also be modified accordingly when the user's intent changes. By invoking new matching templates, the accuracy of template selection is further enhanced based on the original template, enabling optimization based on user feedback and real-time monitoring and adjustment of user intent changes.

[0049] In one embodiment of the present invention, the intention recognition and analysis module further includes: A similarity calculation module is used to obtain the template similarity between the initial matching function template and the newly added matching function template; A difference acquisition module is used to obtain a preset similarity threshold, calculate the difference between the preset similarity threshold and the template similarity, and obtain the template similarity difference; A difference error recognition module is used to compare the template similarity difference with a preset similarity difference threshold to obtain a difference comparison result; Determine whether there is an intention recognition error based on the difference comparison result, and obtain error recognition information; When there is erroneous recognition information, the matching function template is reacquired until there is no intention recognition error.

[0050] The working principle of the above technical solution is as follows: the similarity calculation module is used to obtain the template similarity between the initial matching function template and the newly added matching function template; A difference acquisition module is used to obtain a preset similarity threshold, calculate the difference between the preset similarity threshold and the template similarity, and obtain a template similarity difference; the template similarity difference is used to reflect the difference between the newly added template and the last matched template, and further reflect the accuracy of the last matched template; A difference error recognition module is used to compare the template similarity difference with a preset similarity difference threshold to obtain a difference comparison result; According to the difference comparison result, it is judged whether there is an intention recognition error and the error recognition information is obtained; by comparing the difference with the difference threshold, it can be ensured that the template is corrected when the deviation is too large, and the waste of resources caused by correcting the template for slight differences can be avoided.

[0051] When there is erroneous recognition information, the matching function template is reacquired until there is no intention recognition error.

[0052] The technical effect of the above technical solution is: by calculating the similarity between the newly added matching function template and the initial matching function template, the difference between the two is quantified, thereby enhancing the recognition sensitivity of the consistency between the newly added template and the initial template; The template similarity difference reflects the degree of difference between the newly added template and the initial template, and further reflects the accuracy of the last matched template, ensuring that the system can dynamically evaluate the reliability of the matching result.

[0053] By comparing the template similarity difference with the preset similarity difference threshold, it is determined whether there is an intention recognition error, which improves the sensitivity of deviation recognition and avoids wrong decisions caused by inaccurate templates. It can prevent the deviation from being corrected in time when it is too large, and avoid the waste of resources caused by frequent corrections due to minor differences.

[0054] In one embodiment of the present invention, the difference error identification module includes: When the template similarity difference is greater than the preset similarity difference threshold, it is determined that there is an intention recognition error; when the template similarity difference is less than or equal to the preset similarity difference threshold, it is determined that there is no intention recognition error; When the template similarity difference is greater than a preset similarity difference threshold, obtaining a ratio of the template similarity difference to the preset similarity difference threshold to obtain a matching deviation coefficient; The multiple of the adjustment parameter of the user input processing data is determined based on the matching deviation coefficient.

[0055] The working principle of the above technical solution is: when the template similarity difference is greater than the preset similarity difference threshold, it is determined that there is an intention recognition error; the existence of the error indicates that the error has reached a level that requires correction; When the template similarity difference is less than or equal to the preset similarity difference threshold, it is determined that there is no intention recognition error; the absence of the error means that the error has not reached the level that requires correction.

[0056] When the template similarity difference is greater than a preset similarity difference threshold, obtaining a ratio of the template similarity difference to the preset similarity difference threshold to obtain a matching deviation coefficient; The multiple of the adjustment parameter of the user input processing data is determined according to the matching deviation coefficient. The matching deviation coefficient is obtained to quantify the deviation adjustment.

[0057] For example: The matching similarity between the user input and the initial template is 0.75, and the template similarity difference is 0.25, which exceeds the preset threshold (0.2), and it is determined that there is an intent recognition error.

[0058] The matching deviation coefficient was calculated to be 1.25, quantifying the degree of deviation.

[0059] According to the matching deviation coefficient, the adjustment parameter multiplier is set to 1.25, which significantly improves the adjustment strength of user input processing data.

[0060] Re-matching using the adjusted parameters generates more accurate matching results and provides a more comprehensive solution.

[0061] All threshold values ​​of the present invention can be determined by those skilled in the art based on routine experience in the art.

[0062] The technical effect of the above technical solution is as follows: by comparing the template similarity difference with a preset threshold, it achieves accurate detection of intent recognition errors, ensuring that correction is triggered only when the deviation reaches a level that requires correction. The matching deviation coefficient quantifies the degree of deviation, making the adjustment process more precise and controllable. The parameters of the user input processing data are dynamically adjusted based on the matching deviation coefficient to ensure that the matching results can quickly converge to the accurate value, improving response speed and accuracy. By quantifying the deviation and dynamically adjusting it, the system can avoid unnecessary correction operations, optimize resource utilization efficiency, and improve the robustness and adaptability of the system.

[0063] In one embodiment of the present invention, the matching method includes: Obtaining a scenario database, a scenario event database, and a scenario event function database of a response information database; Obtain user input processing data, and then obtain matching scenario database, matching event database and matching function database, and then obtain user matching database and matching output data; Obtain a preset template database, and then obtain a scene template library, a template event library, and a template function library, and obtain a matching scene template library, a matching event template library, a matching function template library, and a matching function template according to the matching output data; Obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, determine the intent recognition error information, and adjust the parameters, such as Figure 2 shown.

[0064] The working principle of the above technical solution is: obtain the scene database, scene event database and scene event function database of the reply information library; classify the scenes, events and functions of the reply information library through the library management module to obtain classified reply information of scenes, events and functions.

[0065] Obtain user input processing data, and then obtain the matching scene database, matching event database and matching function database, and then obtain the user matching database and matching output data; obtain accurate input data through the input data matching module, match the input data with scenes, events and functions, and input the matched model data; Obtain a preset template database, and then obtain a scene template library, a template event library, and a template function library; obtain a matching scene template library, a matching event template library, a matching function template library, and a matching function template based on the matching output data; perform template matching of scenes, events, and functions of the matching output data through an output template matching module; Acquire the initial matching function module and the newly added matching function module, obtain the template similarity difference, determine the intent recognition error information, and adjust the parameters. The template correction module can obtain the template with the matching error or the template corresponding to the user's intention to change, and correct and adjust the incorrect template.

[0066] The technical effects of the above technical solution are as follows: the device classifies and matches the user input information of the advertiser's consultation in the database, obtains a matching database, trains the intent classification model through the matching database, outputs matching output data, matches the matching output data with templates of scenes, events and functions, calculates the template similarity, determines whether the template output is accurate or whether the user has changed the intent, and corrects the template; The library management module categorizes the response information library by scenario, event, and function, obtaining classified response information for each scenario, event, and function. This enables coarse-to-fine classification management of the response information by scenario, event, and function, improving the flexibility of coarse-grained data processing.

[0067] The input data matching module obtains accurate input data, matches the input data with scenarios, events, and functions, and then inputs the matched model data. By performing multi-dimensional data matching on the input data with scenarios, events, and functions, matching data that closely matches the user input data is obtained and input into the model, achieving accurate extraction of the input data. This improves the accuracy of the model output data from the source, and further enhances the relevance and fit of the output data. The output template matching module is used to match the output data's scenes, events, and functions. By matching the output data's scenes, events, and functions, multi-dimensional template matching can be performed on the output data, ensuring that each part of the obtained template corresponds to the output data, thereby enhancing the accuracy of template matching. The template correction module can identify mismatched templates or the templates that the user intended to change, and then correct and adjust the erroneous templates. Similarity calculation and correction further identify and correct erroneous templates, achieving automated optimized template matching and improving template correction efficiency and accuracy.

[0068] Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the claims and their equivalents, the present invention is intended to include such modifications and variations.

Claims

1. Real-time semantic analysis and dynamic response template matching device for advertiser consultation, characterized by: The device comprises: The library management module is used to obtain the scene database, scene event database and scene event function database of the reply information library; An input data matching module is used to obtain user input processing data, and then obtain a matching scenario database, a matching event database, and a matching function database, and then obtain a user matching database and matching output data; The output template matching module is used to obtain the preset template database, and then obtain the scene template library, template event library and template function library, and obtain the matching scene template library, matching event template library, matching function template library and matching function template according to the matching output data; The template correction module is used to obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, judge the intention recognition error information, and adjust the parameters.

2. The real-time semantic analysis and dynamic reply template matching device for advertiser consultation according to claim 1, characterized in that: The warehouse management module includes: The scenario classification module is used to construct a response information database based on big data information, classify the data scenarios of the response information database, and obtain a scenario database; An event classification module is used to classify data events in the scene database and obtain a scene event database of the scene database; The function classification module is used to perform data function classification on the scene event database to obtain the scene event function database of the scene database.

3. The real-time semantic analysis and dynamic reply template matching device for advertiser consultation according to claim 1 is characterized in that: The input data matching module includes: A pre-training module is used to obtain user input information, perform pre-training processing on the user input information, and obtain user input processing data; A scene matching module, configured to match the user input processing data with a scene database to obtain a matching scene database; An event matching module, configured to match the user input processing data with a scene event database of a matching scene database to obtain a matching event database of the matching scene database; A function matching module, configured to match the user input processing data with a matching scene database, a matching event database, and a matching function database to obtain a matching scene database, a matching event database, and a matching function database; A matching database construction module, configured to generate a user matching database based on the matching scenario database, the matching event database, and the matching function database; The model output module is used to train the intent classification model through the user matching database and obtain matching output data.

4. The real-time semantic analysis and dynamic reply template matching device for advertiser consultation according to claim 3 is characterized in that: The model output module includes: A model training module is used to train the intent classification model through the scene database, the scene event database, and the scene event function database; The data output module is used to input the user matching database into the intent classification model to obtain matching output data.

5. The real-time semantic analysis and dynamic reply template matching device for advertiser consultation according to claim 1 is characterized in that: The output template matching module includes: A scene template acquisition module is used to obtain a preset template database, classify the preset template database into template scenes, and obtain a scene template library; An event template acquisition module, configured to classify the scene template library into template events and obtain a template event library of the scene template library; The function template acquisition module is used to classify the template functions of the template event library and obtain the template function library of the template event library of the scene template library.

6. The device for real-time semantic analysis and dynamic reply template matching for advertiser consultation according to claim 5, characterized in that: The output template matching module also includes: A scene template matching module, configured to match the matching output data with a scene template library to obtain a matching scene template library; An event template acquisition module, configured to match the matching output data with the template event library of the scene template library through the matching scene template library to obtain a matching event template library of the matching scene template library; The function template acquisition module is used to match the matching output data with the template function library of the scene template library, the template event library and the template function library through the matching event template library to obtain the matching function template of the matching scene template library, the matching event template library and the matching function template library.

7. The real-time semantic analysis and dynamic reply template matching device for advertiser consultation according to claim 1, characterized in that: The template correction module includes: The initial filling module is used to retrieve the initial matching function template from the preset template database by matching the output data; Filling variables of the initial matching function template with the matching output data to obtain a target output template; A new matching module is added, which is used to obtain new user input information and obtain new matching output data through the new user input information; Retrieve a new matching function template from the preset template database according to the new matching output data; The intent recognition analysis module is used to calculate template similarity, determine the intent recognition status based on the template similarity, and obtain intent recognition information.

8. The device for real-time semantic analysis and dynamic response template matching for advertiser consultation according to claim 7, characterized in that: The intention recognition and analysis module also includes: A similarity calculation module is used to obtain the template similarity between the initial matching function template and the newly added matching function template; A difference acquisition module is used to obtain a preset similarity threshold, calculate the difference between the preset similarity threshold and the template similarity, and obtain the template similarity difference; A difference error recognition module is used to compare the template similarity difference with a preset similarity difference threshold to obtain a difference comparison result; Determine whether there is an intention recognition error based on the difference comparison result, and obtain error recognition information; When there is erroneous recognition information, the matching function template is reacquired until there is no intention recognition error.

9. The device for real-time semantic analysis and dynamic reply template matching for advertiser consultation according to claim 8, characterized in that: The difference error recognition module includes: When the template similarity difference is greater than the preset similarity difference threshold, it is determined that there is an intent recognition error; When the template similarity difference is less than or equal to the preset similarity difference threshold, it is determined that there is no intention recognition error; When the template similarity difference is greater than a preset similarity difference threshold, obtaining a ratio of the template similarity difference to the preset similarity difference threshold to obtain a matching deviation coefficient; The multiple of the adjustment parameter of the user input processing data is determined based on the matching deviation coefficient.

10. A matching method for implementing the real-time semantic analysis and dynamic reply template matching device of advertiser consultation according to claim 1, characterized in that: The matching method includes: Obtaining a scenario database, a scenario event database, and a scenario event function database of a response information database; Obtain user input processing data, and then obtain matching scenario database, matching event database and matching function database, and then obtain user matching database and matching output data; Obtain a preset template database, and then obtain a scene template library, a template event library, and a template function library, and obtain a matching scene template library, a matching event template library, a matching function template library, and a matching function template according to the matching output data; Obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference, determine the intent recognition error information, and adjust the parameters.

Images