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

Abstract

The application provides a real-time semantic analysis and dynamic reply template matching device for advertisement customer consultation, relates to the technical field of reply template matching, acquires a scene database, a scene event database and a scene event function database of a reply information library; acquires user input processing data, and then acquires a matching scene database, a matching event database and a matching function database, and then acquires a user matching database and matching output data; acquires a preset template database, and then acquires a scene template library, a template event library and a template function library, acquires 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; acquires an initial matching function module and a newly added matching function module, acquires a template similarity difference value, judges intention recognition error information, adjusts parameters, and high-precision template matching and reply of the consultation information are realized through the method and device.

Description

TECHNICAL FIELD

[0001] The application provides an advertisement customer consultation real-time semantic analysis and dynamic reply template matching device, and relates to the technical field of reply template matching. BACKGROUND

[0002] Traditional semantic analysis reply is usually based on comprehensive processing of big data, lacks pertinence, and is prone to misinterpret the intention of customer consultation information, thereby leading to output of incorrect information, and is difficult to discover and adjust in time, and is also often difficult to perceive in time when the user changes the intention; the traditional method has obvious deficiencies in template matching precision, dynamic adjustment capability, intention recognition error detection, resource utilization efficiency and quantitative analysis, and is difficult to meet the needs of modern intelligent systems for accurate matching and efficient optimization. SUMMARY

[0003] The application provides an advertisement customer consultation real-time semantic analysis and dynamic reply template matching device to solve the above problems.

[0004] The advertisement customer consultation real-time semantic analysis and dynamic reply template matching device provided by the application comprises:

[0005] A library management module is configured to obtain a scene database, a scene event database and a scene event function database of a reply information library.

[0006] An input data matching module is configured to obtain user input processing data, and then obtain a matching scene database, a matching event database and a matching function database, and then obtain a user matching database and matching output data.

[0007] An output template matching module is configured to 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.

[0008] A template correction module is configured to obtain an initial matching function module and a new matching function module, obtain a template similarity difference value, judge intention recognition error information, and perform parameter adjustment.

[0009] Further, the library management module comprises:

[0010] A scene classification module is configured to construct a reply information library through big data information, perform data scene classification on the reply information library, and obtain a scene database.

[0011] An event classification module is configured to perform data event classification on the scene database, and obtain a scene event database of the scene database.

[0012] a function classification module, configured to perform data function classification on the scene event database to obtain a scene event function database of the scene event database of the scene database;

[0013] Further, the input data matching module comprises:

[0014] a pre-training module, configured to acquire user input information, and perform pre-training processing on the user input information to obtain user input processing data;

[0015] a scene matching module, configured to perform matching on the user input processing data based on the scene database to obtain a matched scene database;

[0016] an event matching module, configured to perform matching on the user input processing data based on a scene event database of the matched scene database to obtain a matched event database of the matched scene database;

[0017] a function matching module, configured to perform matching on the user input processing data based on a scene event function database of the matched event database of the matched scene database to obtain a matched function database of the matched event database of the matched scene database;

[0018] a matched database construction module, configured to generate a user matched database according to the matched scene database, the matched event database and the matched function database.

[0019] a model output module, configured to train an intent classification model through the user matched database to obtain matched output data.

[0020] Further, the model output module comprises:

[0021] a model training module, configured to train the intent classification model through the scene database, the scene event database and the scene event function database;

[0022] a data output module, configured to input the user matched database into the intent classification model to obtain the matched output data;

[0023] Further, the output template matching module comprises:

[0024] a scene template acquisition module, configured to acquire a preset template database, and perform template scene classification on the preset template database to obtain a scene template library;

[0025] an event template acquisition module, configured to perform template event classification on the scene template library to obtain a template event library of the scene template library;

[0026] The function template acquisition module is configured to classify the template events in the template event library according to functions, and obtain a function template library of the template event library of the scene template library;

[0027] Further, the output template matching module further comprises:

[0028] The scene template matching module is configured to match the matching output data with the scene template library, and obtain a matching scene template library;

[0029] The event template acquisition module is configured to match the matching output data with the matching scene template library and the template event library of the scene template library, and obtain a matching event template library of the matching scene template library;

[0030] The function template acquisition module is configured to match the matching output data with the matching event template library and the function template library of the template event library of the scene template library, and obtain a matching function template of the matching function template library of the matching event template library of the matching scene template library.

[0031] Further, the template deviation correction module comprises:

[0032] The initial filling module is configured to retrieve an initial matching function template from the preset template database according to the matching output data;

[0033] The initial matching function template is filled with variables according to the matching output data, and a target output template is obtained;

[0034] The new matching module is configured to obtain new user input information, and obtain new matching output data according to the new user input information;

[0035] The new matching function template is retrieved from the preset template database according to the new matching output data;

[0036] The intent recognition analysis module is configured to calculate a template similarity, determine an intent recognition state according to the template similarity, and obtain intent recognition information.

[0037] Further, the intent recognition analysis module further comprises:

[0038] The similarity calculation module is configured to obtain a template similarity of the initial matching function template and the new matching function template;

[0039] The difference value acquisition module is configured to obtain a preset similarity threshold, calculate a difference value between the preset similarity threshold and the template similarity, and obtain a template similarity difference value;

[0040] The difference error recognition module is configured to compare the template similarity difference value with a preset similarity difference threshold, and obtain a difference value comparison result;

[0041] determine whether there is an intention recognition error according to the difference comparison result, and obtain error recognition information;

[0042] When there is error recognition information, the matching function template is reacquired until there is no intention recognition error.

[0043] Further, the difference error recognition module comprises:

[0044] When the template similarity difference value is greater than the preset similarity difference threshold value, it is determined that there is an intention recognition error;

[0045] When the template similarity difference value is less than or equal to the preset similarity difference threshold value, it is determined that there is no intention recognition error;

[0046] When the template similarity difference value is greater than the preset similarity difference threshold value, the ratio of the template similarity difference value to the preset similarity difference threshold value is obtained, and a matching deviation coefficient is obtained.

[0047] According to the matching deviation coefficient, the multiple of the adjustment parameter of the user input processing data is determined.

[0048] Further, the matching method comprises:

[0049] Obtain the scene database, scene event database and scene event function database of the reply information library;

[0050] Obtain user input processing data, and then obtain matching scene database, matching event database and matching function database, and then obtain user matching database and matching output data;

[0051] 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;

[0052] Obtain an initial matching function module and a new matching function module, obtain a template similarity difference value, determine intention recognition error information, and adjust parameters.

[0053] The present application has the following advantages: the device classifies and matches the user input information of the advertisement client consultation database, obtains a matching database, trains an intention classification model through the matching database, outputs matching output data, matches the templates of scenes, events and functions through the matching output data, calculates the template similarity, determines whether the template output is accurate or whether the user has changed the intention, and corrects the template.

[0054] The reply information library is classified by the library management module according to scenes, events and functions, and classified reply information of the scenes, events and functions is obtained.

[0055] The input data is matched by the input data matching module, and the input data is matched according to scenes, events and functions, and the model data after matching is input; the input data is matched according to scenes, events and functions in multiple dimensions, and the matching data closely matching the user input data is obtained and input into the model, so that the input data is accurately refined, the accuracy of the model output data is improved from the source, and the relevance and closeness of the output data are further improved.

[0056] The output data is matched by the output template matching module according to scenes, events and functions; the output data is matched according to scenes, events and functions, and the template matching is performed in multiple dimensions, so that each part of the obtained template corresponds to the matching output data, and the accuracy of the template matching is enhanced.

[0057] The template correction module can obtain the matching error template or the template corresponding to the user's change intention, and correct and adjust the error template. The error template is further identified and corrected through similarity calculation and correction, the template automatic optimization matching is realized, and the correction efficiency and correct matching rate of the template are improved. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 Schematic diagram of the real-time semantic analysis and dynamic reply template matching device for advertisement customer consultation;

[0059] Figure 2 Schematic diagram of the real-time semantic analysis and dynamic reply template matching method for advertisement customer consultation. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0061] In one embodiment of the present application, the real-time semantic analysis and dynamic reply template matching device for advertisement customer consultation comprises:

[0062] The library management module is used for obtaining a scene database, a scene event database and a scene event function database of the reply information library.

[0063] The input data matching module is used for obtaining user input processing data, and then obtaining a matching scene database, a matching event database and a matching function database, and then obtaining a user matching database and matching output data.

[0064] The output template matching module is used for obtaining a preset template database, and then obtaining a scene template library, a template event library and a template function library, and obtaining 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.

[0065] The template correction module is used for obtaining an initial matching function module and a new matching function module, obtaining a template similarity difference value, judging intention recognition error information, and adjusting parameters. Figure 1

[0066] The working principle of the above technical solution is that the library management module is used for obtaining a scene database, a scene event database and a scene event function database of a reply information library; the scene, event and function of the reply information library are classified through the library management module to obtain classified reply information of the scene, event and function.

[0067] The input data matching module is used for obtaining user input processing data, and then obtaining a matching scene database, a matching event database and a matching function database, and then obtaining a user matching database and matching output data; accurate input data is obtained through the input data matching module, and the scene, event and function of the input data are matched, and the model data after matching is input;

[0068] The output template matching module is used for obtaining a preset template database, and then obtaining a scene template library, a template event library and a template function library, and obtaining 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.

[0069] The template correction module is used for obtaining an initial matching function module and a new matching function module, obtaining a template similarity difference value, judging intention recognition error information, and adjusting parameters. Through the template correction module, the template with matching errors or the template corresponding to the intention changed by the user can be obtained, and the error template can be corrected and adjusted.

[0070] The technical effect of the above technical solution is that the device classifies and matches the user input information of the advertisement customer consultation through a database, obtains a matching database, trains an intention classification model through the matching database, outputs matching output data, matches the scene, event and function templates through the matching output data, calculates the template similarity, judges whether the template output is accurate or the user changes the intention, and corrects the template. ​

[0071] The reply information library is classified by the library management module according to scenes, events and functions, and classified reply information of the scenes, events and functions is obtained. The classified management of the scenes, events and functions of the reply information is realized from coarse to fine, and the flexibility of coarse granularity of data processing is improved.

[0072] The input data is matched according to scenes, events and functions by the input data matching module, and the matched model data is input. The input data is matched in multiple dimensions according to scenes, events and functions, and matched data closely fitting the input data of the user is obtained and input into the model, so that the accuracy of the input data is improved, and the relevance and fitting of the output data are further improved.

[0073] The output data is matched according to scenes, events and functions by the output template matching module. The matched output data is matched in multiple dimensions according to scenes, events and functions, so that each part of the obtained template corresponds to the matched output data, and the accuracy of the template matching is improved.

[0074] The template correction module can obtain the matched error template or the template corresponding to the intention of the user for change, and correct and adjust the error template. The error template is further identified and corrected through similarity calculation and correction, the automatic optimization matching of the template is realized, and the correction efficiency and correct matching rate of the template are improved.

[0075] In an embodiment of the present application, the library management module comprises:

[0076] The scene classification module is used for constructing the reply information library through big data information, classifying data scenes of the reply information library, and obtaining a scene database.

[0077] The event classification module is used for classifying data events of the scene database, and obtaining a scene event database of the scene database.

[0078] The function classification module is used for classifying data functions of the scene event database, and obtaining a scene event function database of the scene event database of the scene database.

[0079] The working principle of the above technical solution is that the scene classification module is used for constructing the reply information library through big data information, classifying data scenes of the reply information library, and obtaining a scene database.

[0080] The scene database comprises an advertisement recommendation scene, a transaction scene and a transportation scene.

[0081] An event classification module is configured to classify data events of the scene database to obtain a scene event database of the scene database.

[0082] The scene event database includes click analysis events, conversion rate optimization events, and budget allocation events of an advertising recommendation scene, order payment time, refund application time, and commodity consultation events of a transaction scene, and delivery delay events, logistics query events, and package loss events of a transportation event.

[0083] A function classification module is configured to classify data functions of the scene event database to obtain a scene event function database of the scene event database of the scene database.

[0084] The scene event function database includes logistics query, delay notification, and complaint processing data of a transportation event function,

[0085] payment confirmation, refund processing, and order query data of a transaction event function, and click rate analysis, budget optimization, and creative recommendation data of an advertising event function.

[0086] The above technical solution has the following technical effects: by classifying the reply information library according to the scene, a plurality of scene information libraries can be obtained, the targeted reply analysis of each scene is realized, and the accuracy and efficiency of scene recognition are improved.

[0087] By classifying the reply information library according to the event of the scene, a plurality of scene event information libraries can be obtained, the targeted reply analysis of each event of each scene is realized, and the accuracy and efficiency of scene event recognition are improved.

[0088] By classifying the reply information library according to the function of the event of the scene, a plurality of scene event function information libraries can be obtained, the targeted reply analysis of the function of each event of each scene is realized, and the accuracy and efficiency of scene event function recognition are improved.

[0089] An embodiment of the present application, the input data matching module includes:

[0090] A pre-training module is configured to obtain user input information, and pre-train the user input information to obtain user input processing data.

[0091] A scene matching module is configured to match the user input processing data with the scene database to obtain a matching scene database.

[0092] An event matching module is configured to match the user input processing data with a scene event database of the matching scene database to obtain a matching event database of the matching scene database.

[0093] The function matching module is configured to match the user input processing data with a function database of a matching event database of the matching scenario database, to obtain a matching function database of the matching event database of the matching scenario database.

[0094] The matching library construction module is configured to generate a user matching database according to the matching scenario database, the matching event database and the matching function database.

[0095] The model output module is configured to train an intent classification model through the user matching database, to obtain matching output data.

[0096] The pre-training module is configured to obtain user input information, and pre-train the user input information to obtain user input processing data.

[0097] The scenario matching module is configured to match the user input processing data with a scenario database, to obtain a matching scenario database.

[0098] The event matching module is configured to match the user input processing data with a scenario event database of the matching scenario database, to obtain a matching event database of the matching scenario database.

[0099] The function matching module is configured to match the user input processing data with a scenario event function database of the matching event database of the matching scenario database, to obtain a matching function database of the matching event database of the matching scenario database.

[0100] The matching library construction module is configured to generate a user matching database according to the matching scenario database, the matching event database and the matching function database.

[0101] The model output module is configured to train an intent classification model through the user matching database, to obtain matching output data.

[0102] The pre-training includes:

[0103] The pre-training processing of the user input information includes data cleaning, word segmentation, vectorization and the like, to generate user input processing data.

[0104] Specifically, the pre-training processing includes:

[0105] The noise data (such as irrelevant characters, stop words and the like) is removed.

[0106] segmenting text information into meaningful lexical units.

[0107] converting text or image information into numerical vectors (such as through the word embedding model Word2Vec, BERT, or the image feature extraction model ResNet).

[0108] The technical effects of the above technical solutions are: through pre-training processing, the recognition sensitivity of user input information can be enhanced, more efficient data reply matching of user input information is realized, and user input information is more easily recognized;

[0109] By matching the scene database with user input processing data, a scene database that closely matches the reply information of the user input processing data can be obtained, the scene reply of the user input processing data is enhanced, the probability of reply error scene is reduced, and the accuracy of the scene reply of the user input processing data is improved;

[0110] By matching the scene event database of the scene database with 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, the scene event reply of the user input processing data is enhanced, the probability of reply error scene event is reduced, and the accuracy of the scene event reply of the user input processing data is improved;

[0111] By matching the scene event function database of the scene event database of the scene database with user input processing data, a scene event function database of the scene event database of the scene database that closely matches the reply information of the user input processing data can be obtained, the scene event function reply of the user input processing data is enhanced, the probability of reply error scene event function is reduced, and the accuracy of the scene event function reply of the user input processing data is improved;

[0112] The matching output data output by the intent classification model trained by the user matching database has strong pertinence and closeness to the user matching database, so that the output data and the input data have strong relevance in terms of scene, event, and function.

[0113] In an embodiment of the present application, the model output module comprises:

[0114] The model training module is configured to train an intent classification model by using data such as the scene database, the scene event database, and the scene event function database.

[0115] The data output module is configured to input data such as the user matching database into the intent classification model to obtain matching output data.

[0116] 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 through the scene database, the scene event database and the scene event function database adopts accurate data categories and data levels;

[0117] The data output module is used to input the user matching database into the intent classification model to obtain matching output data; the user matching database is analyzed by the intent classification model, and corresponding data is output for accurate data matching analysis.

[0118] Part of the code of the above content includes:

[0119] # Load data

[0120] scenes_df = pd.read_csv('scenes.csv') # Scene database

[0121] events_df = pd.read_csv('events.csv') # Scene event database

[0122] functions_df = pd.read_csv('functions.csv') # Scene event function database

[0123] # Merge data

[0124] training_data = merge_databases(scenes_df, events_df, functions_df)

[0125] # Feature engineering: convert text data to TF-IDF vectors

[0126] vectorizer = TfidfVectorizer(max_features=5000)

[0127] X = vectorizer.fit_transform(training_data['description']) # Assume the description field contains text descriptions

[0128] y = training_data['category'] # Assume the category field is the target category (scene, event, function)

[0129] # Split training and test sets

[0130] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

[0131] # Training an intent classification model (using random forest as an example)

[0132] model = RandomForestClassifier(n_estimators=100, random_state=42)

[0133] model.fit(X_train, y_train)

[0134] # Evaluation Model

[0135] y_pred = model.predict(X_test) print(f"Model accuracy: {accuracy_score(y_test, y_pred)}").

[0136] The technical effect of the above technical solution is as follows: the intent classification model trained by the scene database, scene event database and scene event function database adopts accurate data categories and data levels, and achieves the output of the most accurate output data with less data to train the model;

[0137] By performing data analysis on the user matching database using an intent classification model and outputting corresponding data, accurate data matching analysis is achieved, closely matching the corresponding scenarios, events, and functions, making the model output data more consistent with the scenarios, events, and functions of the user input information.

[0138] In one embodiment of the present invention, the output template matching module includes:

[0139] The scene template acquisition module is used to acquire a preset template database, classify the preset template database into template scenes, and obtain a scene template library.

[0140] The event template acquisition module is used to classify the scene template library into template events and obtain the scene template library's template event library;

[0141] The function template acquisition module is used to classify the template event library into template functions and obtain the template function library of the scene template library.

[0142] The working principle of the technical solution is as follows: a preset template database is acquired, and then a scene template library, a template event library and a template function library are acquired, and a matching scene template library, a matching event template library, a matching function template library and a matching function template are acquired according to matching output data;

[0143] Specifically, the technical solution comprises:

[0144] A scene template acquisition module is configured to acquire a preset template database, perform template scene classification on the preset template database, and obtain a scene template library; the scene template library comprises a plurality of scene template libraries corresponding to a plurality of scenes of a scene database;

[0145] An event template acquisition module is configured to perform template event classification on the scene template library, and obtain a template event library of the scene template library; the template event library comprises a plurality of event template libraries corresponding to a plurality of event databases of the scene database;

[0146] A function template acquisition module is configured to perform template function classification on the template event library, and obtain a template function library of the template event library of the scene template library; the template function library comprises a plurality of function template libraries corresponding to a plurality of function databases of the scene event database of the scene database.

[0147] The technical effect of the technical solution is as follows: the scene template library comprises a plurality of scene template libraries corresponding to a plurality of scenes of a scene database; each preset scene is classified according to a scene template type, template data of each scene is obtained, the accuracy and independence of the scene template are improved, the scheduling efficiency of the scene template is enhanced, the scheduling resources are reduced, the pertinence and accuracy of the scene template supply are enhanced;

[0148] The template event library comprises a plurality of event template libraries corresponding to a plurality of event databases of the scene database; each event of each preset scene is classified according to a scene event template type, template data of each scene event is obtained, the accuracy and independence of the scene event template are improved, the scheduling efficiency of the scene event template is enhanced, the scheduling resources are reduced, the pertinence and accuracy of the scene event template supply are enhanced;

[0149] The template function library comprises a plurality of function template libraries corresponding to a plurality of function databases of the scene event database of the scene database. The template event library comprises a plurality of event template libraries corresponding to a plurality of event databases of the scene database; each function of each event of each preset scene is classified according to a scene event function template type, template data of each scene event function is obtained, the accuracy and independence of the scene event function template are improved, the scheduling efficiency of the scene event function template is enhanced, the scheduling resources are reduced, the pertinence and accuracy of the scene event function template supply are enhanced.

[0150] In one embodiment of the present application, the output template matching module further comprises:

[0151] a scene template matching module, configured to match the matching output data with a scene template library to obtain a matching scene template library;

[0152] an event template obtaining module, configured to match the matching output data with a template event library of the matching scene template library to obtain a matching event template library of the matching scene template library;

[0153] a function template obtaining module, configured to match the matching output data with a template function library of the template event library of the matching scene template library to obtain a matching function template of a matching function template library of the matching event template library of the matching scene template library.

[0154] The working principle of the above technical solution is that the scene template matching module is configured to match the matching output data with a scene template library to obtain a matching scene template library; the matching scene template library comprises a scene template library matched with the matching output data;

[0155] The event template obtaining module is configured to match the matching output data with a template event library of the matching scene template library to obtain a matching event template library of the matching scene template library; the matching event template library comprises a template event library matched with the matching output data in the matching scene template library;

[0156] The function template obtaining module is configured to match the matching output data with a template function library of the template event library of the matching scene template library to obtain a matching function template of a matching function template library of the matching event template library of the matching scene template library; the matching function template comprises a most matching matching function template of a matching function template library matched with the matching output data in the matching event template library in the matching scene template library.

[0157] The technical effect of the above technical solution is that the scene template library data matched with the matching output data is obtained, the matching output data is closely associated with the scene template library on the basis of the association of the user input data and the matching output data, and then the user input data can be associated with the scene template library, thereby realizing the closely associated scene template library for the user input data and the matching output data, greatly enhancing the association of the selection of the scene template of the user input data, and improving the precision, association and intelligence of the template selection.

[0158] The event template library data of the scene matched with the matching output data is acquired, the matching output data is closely associated with the event template library of the scene on the basis that the user input data is associated with the matching output data, and then the user input data can be associated with the event template library of the scene, the closely associated event template library of the scene for the user input data and the matching output data is realized, the associativity of the selection of the event template of the scene of the user input data is greatly enhanced, and the accuracy, the associativity and the intelligence of the template selection are improved.

[0159] The function template library data of the event of the scene matched with the matching output data is acquired, the matching output data is closely associated with the function template library of the event of the scene on the basis that the user input data is associated with the matching output data, and then the user input data can be associated with the function template library of the event of the scene, the closely associated event template library of the scene for the user input data and the matching output data is realized, the associativity of the selection of the function template of the event of the scene of the user input data is greatly enhanced, and the accuracy, the associativity and the intelligence of the template selection are improved.

[0160] In an embodiment of the present application, the template deviation correction module comprises:

[0161] The initial filling module is configured to retrieve an initial matching function template from a preset template database through the matching output data.

[0162] The initial matching function template is filled with variables through the matching output data to obtain a target output template.

[0163] The new matching module is configured to acquire new user input information and acquire new matching output data through the new user input information.

[0164] The new matching function template is retrieved from the preset template database according to the new matching output data.

[0165] The intention recognition analysis module is configured to calculate a template similarity, determine an intention recognition state according to the template similarity, and obtain intention recognition information.

[0166] The working principle of the above technical solution is as follows: the initial filling module is configured to retrieve an initial matching function template from a preset template database through the matching output data.

[0167] The initial matching function template is filled with variables through the matching output data to obtain a target output template.

[0168] The new matching module is used for obtaining new user input information and obtaining new matching output data through the new user input information.

[0169] The new matching function template is used for calibrating and error recognizing the last matching template.

[0170] The intention recognition analysis module is used for calculating the template similarity, judging the intention recognition state according to the template similarity, and obtaining intention recognition information.

[0171] The initial matching function template is the last matching function template obtained by the new matching function template.

[0172] The target output template is a specific complete template obtained through the matching output data.

[0173] The new matching output data is output data matched according to user feedback information.

[0174] The new matching function template is used for calibrating and error recognizing the last matching template.

[0175] In an embodiment of the present application, the intention recognition analysis module further comprises:

[0176] The similarity calculation module is used for obtaining the template similarity of the initial matching function template and the new matching function template.

[0177] The difference obtaining module is used for obtaining a preset similarity threshold, calculating the difference between the preset similarity threshold and the template similarity, and obtaining a template similarity difference.

[0178] The difference error recognizing module is used for comparing the template similarity difference with a preset similarity difference threshold, and obtaining a difference comparison result.

[0179] determine whether there is an intention recognition error according to the difference comparison result, and obtain error recognition information;

[0180] When there is error recognition information, the matching function template is reacquired until there is no intention recognition error.

[0181] The working principle of the above technical solution is that the similarity calculation module is used to obtain the template similarity of the initial matching function template and the newly added matching function template.

[0182] The 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.

[0183] The difference error recognition module is used to compare the template similarity difference with a preset similarity difference threshold, and obtain a difference comparison result.

[0184] According to the difference comparison result, it is determined whether there is an intention recognition error, and error recognition information is obtained.

[0185] When there is error recognition information, the matching function template is reacquired until there is no intention recognition error.

[0186] The technical effect of the above technical solution is that by calculating the similarity of the newly added matching function template and the initial matching function template, the difference between the two is quantified, and the recognition sensitivity of the consistency of the newly added template and the initial template is enhanced.

[0187] The template similarity difference reflects the difference between the newly added template and the initial template, and further reflects the accuracy of the last matched template, so as to ensure that the system can dynamically evaluate the reliability of the matching result.

[0188] By comparing the template similarity difference with the preset similarity difference threshold, it is determined whether there is an intention recognition error, the sensitivity of deviation recognition is improved, error decision caused by inaccurate templates is avoided, deviation that is too large is prevented from not being corrected in time, and resource waste caused by frequent correction due to slight differences is avoided.

[0189] In an embodiment of the present application, the difference error recognition module comprises:

[0190] When the template similarity difference value is greater than the preset similarity difference threshold value, it is determined that there is an intent recognition error.

[0191] When the template similarity difference value is greater than the preset similarity difference threshold value, the ratio of the template similarity difference value to the preset similarity difference threshold value is obtained to obtain a matching deviation coefficient.

[0192] The multiple of the adjustment parameter of the user input processing data is determined according to the matching deviation coefficient.

[0193] The working principle of the above technical solution is that when the template similarity difference value is greater than the preset similarity difference threshold value, it is determined that there is an intent recognition error; the existence represents that the correction is needed.

[0194] When the template similarity difference value is less than or equal to the preset similarity difference threshold value, it is determined that there is no intent recognition error; the nonexistence represents that the correction is not needed.

[0195] When the template similarity difference value is greater than the preset similarity difference threshold value, the ratio of the template similarity difference value to the preset similarity difference threshold value is obtained to obtain a matching deviation coefficient.

[0196] The multiple of the adjustment parameter of the user input processing data is determined according to the matching deviation coefficient. The matching deviation coefficient quantifies the deviation adjustment.

[0197] For example:

[0198] The matching similarity of the user input and the initial template is 0.75, the template similarity difference value is 0.25, which exceeds the preset threshold value (0.2), and it is determined that there is an intent recognition error.

[0199] The matching deviation coefficient is calculated as 1.25, which quantifies the degree of deviation.

[0200] According to the matching deviation coefficient, the adjustment parameter multiple is set to 1.25, which significantly improves the adjustment strength of the user input processing data.

[0201] Using the adjusted parameter to rematch, a more accurate matching result is generated, and a more comprehensive solution is provided.

[0202] All threshold values of the present application can be determined by those skilled in the art according to the conventional experience in the art.

[0203] The technical effects of the above technical solutions are: by comparing the template similarity difference value with the preset threshold value, accurate intent recognition error detection is realized, and it is ensured that correction is triggered only when the deviation reaches a degree that needs to be corrected. The matching deviation coefficient quantifies the degree of deviation, making the adjustment process more accurate and controllable. According to the matching deviation coefficient, the parameters of the user input processing data are dynamically adjusted, ensuring that the matching result can quickly converge to the accurate value, improving the response speed and accuracy. Through quantitative deviation and dynamic adjustment, the system can avoid unnecessary correction operations, optimize resource utilization efficiency, and improve the robustness and adaptability of the system.

[0204] In an embodiment of the present application, the matching method comprises:

[0205] Obtain the scene database, scene event database and scene event function database of the reply information library;

[0206] Obtain the 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;

[0207] 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;

[0208] Obtain the initial matching function module and the newly added matching function module, obtain the template similarity difference value, judge the intent recognition error information, and adjust the parameters, such as Figure 2 as shown.

[0209] The working principle of the above technical solutions is: obtaining the scene database, scene event database and scene event function database of the reply information library; classifying the scene, event and function of the reply information library through the library management module to obtain the classified reply information of the scene, event and function.

[0210] Obtain the 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 scene, event and function of the input data, and input the matched model data;

[0211] 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; perform template matching of the scene, event and function of the matching output data through the output template matching module;

[0212] The system includes modules for initial matching and new matching, which retrieve template similarity differences, identify intent recognition errors, and adjust parameters. The template correction module retrieves incorrectly matched templates or templates corresponding to user-mandated changes, allowing for the correction and adjustment of these templates.

[0213] The technical effects of the above technical solution are as follows: This device classifies and matches user input information from advertising clients into a database, obtains a matching database, trains an intent classification model through the matching database, outputs matching output data, matches templates for scenarios, events, and functions through the matching output data, calculates template similarity, and judges whether the template output is accurate or whether the user has changed their intent, and corrects the template accordingly.

[0214] The database management module categorizes response information into scenarios, events, and functions, obtaining categorized response information for each scenario, event, and function. This achieves coarse-to-fine categorization management of response information by scenario, event, and function, improving the flexibility of data processing at the coarse-grained level.

[0215] The input data matching module obtains accurate input data, matches the input data with scenarios, events, and functions, and inputs the matched model data. By performing multi-dimensional data matching of scenarios, events, and functions, matching data that closely matches the user input data is obtained and input into the model, realizing the precise extraction of input data and improving the accuracy of the model output data from the source, further improving the relevance and fit of the output data.

[0216] The output template matching module matches the scenarios, events, and functions of the output data. This allows for multi-dimensional template matching of the output data, ensuring that every part of the resulting template corresponds to the matched output data, thus enhancing the accuracy of template matching.

[0217] The template correction module can identify and adjust incorrect templates or templates corresponding to user intent to change. Through similarity calculation and correction, it further identifies and corrects incorrect templates, achieving automated template optimization and improving template correction efficiency and accuracy.

[0218] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A device for real-time semantic analysis and dynamic reply template matching of an advertiser's inquiry, characterized in that, The device comprises: A library management module for obtaining a scenario database, a scenario event database and a scenario event function database of a reply information library; An input data matching module for obtaining user input processing data, and then obtaining a matching scenario database, a matching event database and a matching function database, and then obtaining a user matching database and matching output data; An output template matching module for obtaining a preset template database, and then obtaining a scenario template library, a template event library and a template function library, and obtaining a matching scenario template library, a matching event template library, a matching function template library and a matching function template according to the matching output data; A template correction module for obtaining an initial matching function template and a new matching function template, obtaining a template similarity difference value, judging intention recognition error information and performing parameter adjustment. The library management module comprises: A scenario classification module for constructing a reply information library through big data information, performing data scenario classification on the reply information library, and obtaining a scenario database; An event classification module for performing data event classification on the scenario database, and obtaining a scenario event database of the scenario database; A function classification module for performing data function classification on the scenario event database, and obtaining a scenario event function database of the scenario event database of the scenario database. The input data matching module comprises: A pre-training module for obtaining user input information, performing pre-training processing on the user input information, and obtaining user input processing data; A scenario matching module for matching the user input processing data with the scenario database, and obtaining a matching scenario database; An event matching module for matching the user input processing data with the scenario event database of the matching scenario database, and obtaining a matching event database of the matching scenario database; A function matching module for matching the user input processing data with the scenario event function database of the matching event database of the matching scenario database, and obtaining a matching function database of the matching event database of the matching scenario database; A matching library construction module for generating a user matching database according to the matching scenario database, the matching event database and the matching function database; A model output module for training an intention classification model through the user matching database, and obtaining matching output data; The output template matching module comprises: A scenario template acquisition module for obtaining a preset template database, performing template scenario classification on the preset template database, and obtaining a scenario template library; An event template acquisition module for performing template event classification on the scenario template library, and obtaining a template event library of the scenario template library; A function template acquisition module for performing template function classification on the template event library, and obtaining a template function library of the template event library of the scenario template library.

2. The apparatus for real-time semantic analysis of a client inquiry and dynamic reply template matching of claim 1, wherein, The model output module comprises: A model training module for training an intention classification model through the scenario database, the scenario event database and the scenario event function database; A data output module for inputting the user matching database into the intention classification model, and obtaining matching output data.

3. The apparatus for real-time semantic analysis and dynamic reply template matching of advertiser inquiries according to claim 1, wherein, The output template matching module further comprises: The scene template matching module is configured to match the matching output data with a scene template library to obtain a matching scene template library. The event template obtaining module is configured to match the matching output data with a template event library of the matching scene template library to obtain a matching event template library of the matching scene template library. The function template obtaining module is configured to match the matching output data with a template function library of the template event library of the matching scene template library to obtain a matching function template of the matching function template library of the matching event template library of the matching scene template library.

4. The apparatus for real-time semantic analysis and dynamic reply template matching of advertiser inquiries according to claim 1, wherein, The template deviation correction module comprises: The initial filling module is configured to call an initial matching function template from the preset template database according to the matching output data. The initial matching function template is filled with variables according to the matching output data to obtain a target output template. The new matching module is configured to obtain new user input information and obtain new matching output data according to the new user input information. The new matching function template is called from the preset template database according to the new matching output data. The intent recognition analysis module is configured to calculate a template similarity, determine an intent recognition state according to the template similarity, and obtain intent recognition information.

5. The apparatus for real-time semantic analysis of a client query and matching to a dynamic reply template of claim 4, wherein, The intent recognition analysis module further comprises: The similarity calculation module is configured to obtain a template similarity of the initial matching function template and the new matching function template. The difference obtaining module is configured to obtain a preset similarity threshold, calculate a difference between the preset similarity threshold and the template similarity, and obtain a template similarity difference. The difference error recognition module is configured to compare the template similarity difference with a preset similarity difference threshold to obtain a difference comparison result. It is determined whether there is an intent recognition error according to the difference comparison result, and error recognition information is obtained. When there is error recognition information, the matching function template is reacquired until there is no intent recognition error.

6. The apparatus for real-time semantic analysis of a client query and matching to a dynamic reply template of claim 5, wherein, The difference error recognition module comprises: 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 intent recognition error. When the template similarity difference is greater than the preset similarity difference threshold, a ratio of the template similarity difference to the preset similarity difference threshold is obtained to obtain a matching deviation coefficient. The matching method comprises:

7. A matching method for the apparatus for implementing real-time semantic analysis and dynamic reply template matching of the advertiser consultation according to claim 1, characterized in that, Obtain a scene database, a scene event database, and a scene event function database of a reply information library. Obtain user input processing data, and then obtain a matching scene database, a matching event database, and a matching function database, and then obtain a 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 an initial matching function template and a new matching function template, obtain a template similarity difference value, judge intent recognition error information, and perform parameter adjustment; The scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply information library are obtained, and the scene database, the scene event database and the scene event function database of the reply ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Images