Car purchase intention recognition system based on ERNIE model

Abstract

The invention relates to the technical field of electronic data processing, in particular to a car purchase intention recognition system based on an ERNIE model, and the system comprises a data collection module which is used for collecting the natural language data of a user; the data optimization module is used for optimizing the natural language data of the user; the data preprocessing module is used for cleaning the optimized user data and dividing the preprocessed ERNIE model data into a training set, a verification set and a test set; the data model fine tuning module is used for performing fine tuning on the initial ERNIE model to obtain a fine-tuned ERNIE model; the model perfection module is used for performing model perfection on the fine-tuning ERNIE model to obtain a car purchase intention recognition model; and the car purchase intention recognition module is used for inputting the user natural language data collected in real time into the car purchase intention recognition model. According to the invention, the accuracy of successful identification of the car purchase intention of the user in the second-hand car transaction process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic data processing, and in particular to a car purchase intention recognition system based on an ERNIE model. Background Art

[0002] With the development of digital technology and the popularization of the Internet, the used car market has undergone a transformation from traditional offline transactions to online platform transactions. In this process, a large amount of user behavior data and communication information are generated, providing companies with opportunities to deeply understand customer needs and optimize services. At present, the used car trading platforms on the market mainly rely on data directly input by users to capture customer intentions and recommend products. They are unable to fully capture users' implicit needs and deeper purchasing intentions, resulting in insufficient service efficiency and user satisfaction. Currently, the widely used technologies mainly include keyword search and basic data analysis methods. For example, by analyzing the keywords entered by users to infer user intentions, or by statistically analyzing users' click behaviors to infer their preferences. These methods often rely on more intuitive data interpretation and lack an understanding of the deeper intentions behind user language and behavior.

[0003] Chinese patent publication number: CN117237006A discloses a vehicle marketing method, device, equipment and storage medium; the method includes: obtaining structured data obtained by the salesperson communicating with the first user who wants to buy the vehicle and unstructured data of the first user during the test drive; based on the structured data, the unstructured data, the vehicle purchase knowledge graph and the vehicle purchase decision model, determining the vehicle purchase intention of the first user; based on the vehicle purchase intention, determining the vehicle marketing strategy. However, the solution does not perform data analysis and processing on the user's language information, and it is difficult to improve the accuracy of successfully identifying the user's vehicle purchase intention during the second-hand car transaction process. Summary of the invention

[0004] To this end, the present invention provides a car purchase intention recognition system based on the ERNIE model to overcome the problem of low accuracy in successfully recognizing user car purchase intentions during second-hand car transactions in the prior art.

[0005] To achieve the above object, the present invention provides a car purchase intention recognition system based on the ERNIE model, characterized in that the system comprises:

[0006] A data collection module is used to collect user natural language data;

[0007] A data optimization module is used to optimize the user's natural language data to obtain optimized user data;

[0008] The data preprocessing module is used to clean the optimized user data to obtain preprocessed ERNIE model data, and divide the preprocessed ERNIE model data into a training set, a validation set, and a test set;

[0009] A data model fine-tuning module is used to fine-tune the preliminary ERNIE model to obtain a fine-tuned ERNIE model;

[0010] The model improvement module is used to improve the fine-tuned ERNIE model to obtain a car purchase intention recognition model;

[0011] The car purchase intention recognition module is used to input the user's natural language data collected in real time into the car purchase intention recognition model and output the customer intention prediction data.

[0012] Furthermore, the data optimization module includes:

[0013] A data stream processing unit is used to process the user's natural language data in real time to obtain instant and continuous user language data;

[0014] A data verification and cleaning unit is used to verify and repair the real-time continuous user language data to obtain high-quality user language data;

[0015] An adaptive data collection unit, used to adjust the collection frequency of high-quality user language data in real time to obtain sufficient high-quality user language data;

[0016] An intelligent error processing unit is used to automatically identify errors in the sufficient and high-quality user language data and repair the errors, and use the repaired sufficient and high-quality user language data as optimized user data.

[0017] Furthermore, the data stream processing unit uses the data stream processing technology Apache Kafka to process the user's natural language data in real time to obtain instant and continuous user language data;

[0018] The data verification and cleaning unit uses a text recognition processing tool to verify and repair the format and content of the instant and continuous user language data to obtain high-quality user language data.

[0019] Furthermore, the adaptive data collection unit uses an API tool to adjust the frequency and range of high-quality user language data collection to obtain sufficient high-quality user language data;

[0020] The intelligent error processing unit uses the Informatica Data Quality tool and applies the error repair strategy in the Informatica Data Quality tool to repair the errors that occur during the collection of the sufficient and high-quality user language data, thereby obtaining optimized user data.

[0021] Furthermore, the data preprocessing module includes:

[0022] A structure correction unit, used for correcting the grammatical structure of the optimized user data to obtain corrected user data;

[0023] The de-duplication unit is used to remove personal duplicate data in the corrected user data to obtain pre-processed ERNIE model data, and divide the pre-processed ERNIE model data into a training set and a test set.

[0024] Furthermore, the structure correction unit uses a text data repair tool to perform vocabulary correction, grammar correction and structure correction on the optimized user data to obtain corrected user data;

[0025] The de-duplication unit performs de-duplication processing on the corrected user data through a data processing tool to obtain pre-processed ERNIE model data, and divides the pre-processed ERNIE model data into 70% training set and 30% test set.

[0026] Furthermore, the data model fine-tuning module sets the learning rate P of the pre-trained ERNIE model to P=P0 / 10, sets P0 as the original learning rate, and obtains the fine-tuned ERNIE model.

[0027] Furthermore, the model improvement module includes:

[0028] The intent recognition training unit is used to train the fine-tuned ERNIE model according to the training set, input the verification set into the trained fine-tuned ERNIE model, and output the verification accuracy K;

[0029] The model optimization unit is used to compare the verification accuracy K with the preset accuracy K0, and judge the practicality of the fine-tuning ERNIE model according to the comparison result, and output the fine-tuning ERNIE model according to the judgment result.

[0030] Furthermore, the intention recognition training unit inputs the training set into the fine-tuned ERNIE model for training, inputs the verification set into the trained fine-tuned ERNIE model, and outputs the verification accuracy K, wherein the number of correct verifications is set to Q, the number of verification set samples is Q0, and the verification accuracy K = Q / Q0.

[0031] Furthermore, the model optimization unit compares the verification accuracy K with the preset accuracy K0, and judges the practicality of the trained fine-tuning ERNIE model according to the comparison result, and outputs the fine-tuning ERNIE model according to the judgment result, wherein:

[0032] When K≥K0, the trained fine-tuned ERNIE model is judged to be of practicality standard, and the trained fine-tuned ERNIE model is output as a car purchase intention recognition model;

[0033] When K<K0, it is determined that the practicality of the trained fine-tuned ERNIE model does not meet the standard, and the data collection module is returned to obtain the natural language data of the second user, and the fine-tuned ERNIE model is trained according to the natural language data of the second user until the practicality of the trained fine-tuned ERNIE model meets the standard.

[0034] Compared with the prior art, the beneficial effect of the present invention lies in that the system collects user natural language data through a data collection module, and then performs preliminary screening and optimization on the user natural language data through a data optimization module to obtain optimized user data, and then performs noise removal, grammatical structure correction and duplicate data removal on the optimized user data through a data preprocessing module to obtain preprocessed ERNIE model data, and the preprocessed ERNIE model data is input into a fine-tuned ERNIE model generated by a data model fine-tuning module to generate a preliminary prediction result of the user's car purchase intention, and the model is further adjusted according to the prediction accuracy of the user's car purchase intention through a model improvement module to obtain a model for identifying the user's car purchase intention based on the ERNIE model, thereby improving the efficiency of analyzing the user's car purchase intention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the structure of the car purchase intention recognition system based on the ERNIE model in this embodiment;

[0036] Figure 2 This is a schematic diagram of the structure of the data optimization module of this embodiment;

[0037] Figure 3 Schematic diagram of the structure of the data preprocessing module of this embodiment;

[0038] Figure 4 This is a schematic diagram of the structure of the model improvement module of this embodiment. DETAILED DESCRIPTION

[0039] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0041] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0042] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0043] See also Figure 1 As shown, it is a structural schematic diagram of the car purchase intention recognition system based on the ERNIE model in this embodiment, and the system includes:

[0044] A data collection module is used to collect user natural language data;

[0045] A data optimization module is used to optimize the user's natural language data to obtain optimized user data. The data optimization module is connected to the data collection module;

[0046] The data preprocessing module is used to clean the optimized user data to obtain preprocessed ERNIE model data, and divide the preprocessed ERNIE model data into a training set, a validation set, and a test set. The data preprocessing module is connected to the data optimization module;

[0047] A data model fine-tuning module is used to fine-tune the preliminary ERNIE model to obtain a fine-tuned ERNIE model. The data model fine-tuning module is connected to the data preprocessing module.

[0048] A model improvement module is used to improve the fine-tuned ERNIE model to obtain a car purchase intention recognition model. The model improvement module is connected to the data model fine-tuning module;

[0049] The car purchase intention recognition module is used to input the user's natural language data collected in real time into the car purchase intention recognition model and output the customer intention prediction data. The car purchase intention recognition module is connected to the model improvement module.

[0050] Specifically, the system is set in a user's car purchase terminal, predicts the user's car purchase intention through the ERNIE model, and obtains a more accurate prediction system through continuous optimization of the model results, thereby improving the analysis efficiency of the user's car purchase intention. The system collects the user's natural language data through a data collection module, and then preliminarily screens and optimizes the user's natural language data through a data optimization module to obtain optimized user data, and then removes noise words, grammatical structure corrections and duplicate data from the optimized user data through a data preprocessing module to obtain preprocessed ERNIE model data, and the preprocessed ERNIE model data is input into a fine-tuned ERNIE model generated by a data model fine-tuning module to generate a preliminary prediction result of the user's car purchase intention, and the model is further adjusted according to the accuracy of the user's car purchase intention prediction through a model improvement module to obtain a model for identifying the user's car purchase intention based on the ERNIE model, thereby improving the analysis efficiency of the user's car purchase intention.

[0051] Specifically, the user natural language data refers to the text records generated when the user uses online social media and trading platforms. This embodiment does not limit the method for the data collection module to collect user natural language data. Relevant technical personnel in this field can freely set it according to actual needs. For example, the data collection module can collect user natural language data through online social media and trading platforms.

[0052] See also Figure 2 As shown, it is a schematic diagram of the structure of the data optimization module of this embodiment, and the data optimization module includes:

[0053] A data stream processing unit is used to process the user's natural language data in real time to obtain instant and continuous user language data;

[0054] A data verification and cleaning unit is used to verify and repair the instant and continuous user language data to obtain high-quality user language data. The data verification and cleaning unit is connected to the data stream processing unit;

[0055] An adaptive data collection unit, used to adjust the collection frequency of high-quality user language data in real time to obtain sufficient high-quality user language data, and the adaptive data collection unit is connected to the data verification and cleaning unit;

[0056] The intelligent error processing unit is used to automatically identify errors in the sufficient and high-quality user language data and repair the errors, and use the repaired sufficient and high-quality user language data as optimized user data. The intelligent error processing unit is connected to the adaptive data acquisition unit.

[0057] Specifically, the data stream processing unit uses the data stream processing technology Apache Kafka to process the user's natural language data in real time to obtain instant and continuous user language data.

[0058] Specifically, the data stream processing unit processes the user's natural language data in real time to improve the real-time performance of data processing, thereby ensuring information immediacy and data continuity when collecting data.

[0059] Specifically, the data stream processing technology Apache Kafka refers to an open source messaging system that performs unified, high-throughput, and low-wait processing on real-time data.

[0060] Specifically, the data verification and cleaning unit uses a text recognition processing tool to verify and repair the format and content of the instant and continuous user language data to obtain high-quality user language data.

[0061] Specifically, the data verification and cleaning unit processes the immediate and continuous user language data in real time to improve the accuracy and completeness of the data.

[0062] Specifically, this embodiment does not limit the text recognition and processing tools. Relevant personnel can set them at will as long as they meet the text recognition and processing requirements. For example, the OCR.Space tool can be used. The OCR.Space tool is a powerful online character recognition tool with functions such as data format and content verification, removal of duplicate records, correction of structural errors, and filling of missing values.

[0063] Specifically, the adaptive data collection unit uses an API tool to adjust the frequency and range of collecting high-quality user language data to obtain sufficient high-quality user language data.

[0064] Specifically, the adaptive data collection unit will use the API tool to increase the collection frequency for high-frequency accessed data points to provide more accurate data feedback. The API tool has the function of adjusting the frequency and range of data collection.

[0065] Specifically, the intelligent error processing unit uses the Informatica Data Quality tool and applies the error repair strategy in the Informatica Data Quality tool to repair the errors that occur during the collection of the sufficient and high-quality user language data, thereby obtaining optimized user data.

[0066] Specifically, the Informatica Data Quality tool refers to a data quality management tool with data analysis, data cleaning, data matching and data verification functions.

[0067] See also Figure 3 As shown, it is a schematic diagram of the structure of the data preprocessing module of this embodiment, and the data preprocessing module includes:

[0068] A structure correction unit, used for correcting the grammatical structure of the optimized user data to obtain corrected user data;

[0069] The de-duplication unit is used to remove personal duplicate data in the corrected user data to obtain pre-processed ERNIE model data, and divide the pre-processed ERNIE model data into a training set and a test set. The de-duplication unit is connected to the structure correction unit.

[0070] Specifically, the structure correction unit uses a text data repair tool to perform vocabulary correction, grammar correction and structure correction on the optimized user data to obtain corrected user data.

[0071] Specifically, the vocabulary correction refers to the correction of typos, inappropriate words, and confusing words. The grammar correction refers to the correction of tense, voice, subject-verb agreement, missing sentence components, and redundancy. The structure correction refers to the correction of the rationality of sentence structure, the logic of paragraphs, and the overall layout of chapters. This embodiment does not limit the text data repair tool. Relevant technicians in this field can freely set it according to actual needs. It only needs to meet the vocabulary correction, grammar correction, and structure correction functions. For example, Grammarly can be used. Grammarly is a software with text data repair function, which can mark vocabulary, grammar, and structure errors and give correction suggestions.

[0072] Specifically, the deduplication unit performs deduplication processing on the corrected user data through a data processing tool to obtain pre-processed ERNIE model data, and divides the pre-processed ERNIE model data into 70% training set and 30% test set.

[0073] Specifically, the deduplication processing refers to the deletion of repeated text data according to needs, the training set refers to the data set used to train the ERNIE model, and the test set refers to the data set used to evaluate the performance of the ERNIE model. This embodiment does not limit the type of data processing tool. Relevant technical personnel in this field can freely set it according to actual needs, and it is only necessary to meet the needs of deduplication processing of corrected user data. For example, the type of data processing tool can be set to MySQL, and MySQL refers to a relational database management system that performs pre-processing such as formatting, denoising and cleaning of data.

[0074] Specifically, the data model fine-tuning module sets the learning rate P of the pre-trained ERNIE model to P=P0 / 10, sets P0 as the original learning rate, and obtains the fine-tuned ERNIE model.

[0075] Specifically, the pre-trained ERNIE model is a knowledge-based natural language representation model with certain semantic recognition and representation functions. The learning rate is a hyperparameter in the deep learning model training, which determines the speed of model parameter adjustment. The original learning rate refers to the original learning rate of the pre-trained ERNIE model. This embodiment does not limit the specific value of the original learning rate P0. Relevant technical personnel in this field can freely set it according to actual needs, as long as the needs of the pre-trained ERNIE model are met, such as setting P0=0.01.

[0076] Specifically, in the data model fine-tuning module, the learning rate P of the pre-trained ERNIE model is set to one tenth of the original learning rate to avoid destroying the original useful data of the model.

[0077] See also Figure 4 As shown, it is a schematic diagram of the structure of the model improvement module of this embodiment, and the model improvement module includes:

[0078] The intent recognition training unit is used to train the fine-tuned ERNIE model according to the training set, input the verification set into the trained fine-tuned ERNIE model, and output the verification accuracy K;

[0079] The model optimization unit is used to compare the verification accuracy K with the preset accuracy K0, and judge the practicality of the fine-tuned ERNIE model according to the comparison result, and output the fine-tuned ERNIE model according to the judgment result. The model optimization unit is connected to the intent recognition training unit.

[0080] Specifically, the intention recognition training unit inputs the training set into the fine-tuned ERNIE model for training, inputs the verification set into the trained fine-tuned ERNIE model, and outputs the verification accuracy K, wherein the number of correct verifications is set to Q, the number of verification set samples is Q0, and the verification accuracy K = Q / Q0.

[0081] Specifically, the model optimization unit compares the verification accuracy K with the preset accuracy K0, and judges the practicality of the trained fine-tuning ERNIE model according to the comparison result, and outputs the fine-tuning ERNIE model according to the judgment result, wherein:

[0082] When K≥K0, the trained fine-tuned ERNIE model is judged to be of practicality standard, and the trained fine-tuned ERNIE model is output as a car purchase intention recognition model;

[0083] When K<K0, it is determined that the practicality of the trained fine-tuned ERNIE model does not meet the standard, and the data collection module is returned to obtain the natural language data of the second user, and the fine-tuned ERNIE model is trained according to the natural language data of the second user until the practicality of the trained fine-tuned ERNIE model meets the standard.

[0084] Specifically, the correct verification number Q refers to the number of samples correctly identified by the trained fine-tuned ERNIE model for the verification set, and the preset accuracy rate K0 refers to a preset value reflecting the practicality of the fine-tuned ERNIE model. This embodiment does not limit the value of the preset accuracy rate K0, and relevant technical personnel in the field can freely set it according to actual needs, as long as it meets the requirement of reflecting the practicality of the fine-tuned ERNIE model. For example, the preset accuracy rate K0 can be set to 0.8.

[0085] Specifically, the car purchase intention recognition module inputs the user's natural language data collected in real time into the car purchase intention recognition model, and outputs customer intention prediction data.

[0086] Specifically, the customer intention prediction data includes whether the user has the intention to buy a car and whether the user has no intention to buy a car.

[0087] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A car purchase intention recognition system based on the ERNIE model, characterized in that: The system comprises: A data collection module is used to collect user natural language data; A data optimization module is used to optimize the user's natural language data to obtain optimized user data; The data preprocessing module is used to clean the optimized user data to obtain preprocessed ERNIE model data, and divide the preprocessed ERNIE model data into a training set, a validation set, and a test set; A data model fine-tuning module is used to fine-tune the preliminary ERNIE model to obtain a fine-tuned ERNIE model; The model improvement module is used to improve the fine-tuned ERNIE model to obtain a car purchase intention recognition model; The car purchase intention recognition module is used to input the user's natural language data collected in real time into the car purchase intention recognition model and output the customer intention prediction data.

2. The car purchase intention recognition system based on the ERNIE model according to claim 1 is characterized in that: The data optimization module includes: A data stream processing unit is used to process the user's natural language data in real time to obtain instant and continuous user language data; A data verification and cleaning unit is used to verify and repair the real-time continuous user language data to obtain high-quality user language data; An adaptive data collection unit, used to adjust the collection frequency of high-quality user language data in real time to obtain sufficient high-quality user language data; An intelligent error processing unit is used to automatically identify errors in the sufficient and high-quality user language data and repair the errors, and use the repaired sufficient and high-quality user language data as optimized user data.

3. The car purchase intention recognition system based on the ERNIE model according to claim 2 is characterized in that: The data stream processing unit uses the data stream processing technology Apache Kafka to process the user's natural language data in real time to obtain instant and continuous user language data; The data verification and cleaning unit uses a text recognition processing tool to verify and repair the format and content of the instant and continuous user language data to obtain high-quality user language data.

4. The car purchase intention recognition system based on the ERNIE model according to claim 3 is characterized in that: The adaptive data collection unit uses an API tool to adjust the frequency and range of high-quality user language data collection to obtain sufficient high-quality user language data; The intelligent error processing unit uses the Informatica Data Quality tool and applies the error repair strategy in the Informatica Data Quality tool to repair the errors that occur during the collection of the sufficient and high-quality user language data, thereby obtaining optimized user data.

5. The car purchase intention recognition system based on the ERNIE model according to claim 1 is characterized in that: The data preprocessing module comprises: A structure correction unit, used for correcting the grammatical structure of the optimized user data to obtain corrected user data; The de-duplication unit is used to remove personal duplicate data in the corrected user data to obtain pre-processed ERNIE model data, and divide the pre-processed ERNIE model data into a training set and a test set.

6. The car purchase intention recognition system based on the ERNIE model according to claim 5 is characterized in that: The structure correction unit uses a text data repair tool to perform vocabulary correction, grammar correction and structure correction on the optimized user data to obtain corrected user data; The de-duplication unit performs de-duplication processing on the corrected user data through a data processing tool to obtain pre-processed ERNIE model data, and divides the pre-processed ERNIE model data into 70% training set and 30% test set.

7. The car purchase intention recognition system based on the ERNIE model according to claim 1 is characterized in that: The data model fine-tuning module sets the learning rate P of the pre-trained ERNIE model to P=P0 / 10, sets P0 as the original learning rate, and obtains the fine-tuned ERNIE model.

8. The car purchase intention recognition system based on the ERNIE model according to claim 1 is characterized in that: The model improvement module includes: The intent recognition training unit is used to train the fine-tuned ERNIE model according to the training set, input the verification set into the trained fine-tuned ERNIE model, and output the verification accuracy K; The model optimization unit is used to compare the verification accuracy K with the preset accuracy K0, and judge the practicality of the fine-tuning ERNIE model according to the comparison result, and output the fine-tuning ERNIE model according to the judgment result.

9. The car purchase intention recognition system based on the ERNIE model according to claim 8 is characterized in that: The intention recognition training unit inputs the training set into the fine-tuned ERNIE model for training, inputs the verification set into the trained fine-tuned ERNIE model, and outputs the verification accuracy K, wherein the number of correct verifications is set to Q, the number of verification set samples is Q0, and the verification accuracy K = Q / Q0.

10. The car purchase intention recognition system based on the ERNIE model according to claim 9 is characterized in that: The model optimization unit compares the verification accuracy K with the preset accuracy K0, and judges the practicality of the trained fine-tuning ERNIE model according to the comparison result, and outputs the fine-tuning ERNIE model according to the judgment result, wherein: When K≥K0, the trained fine-tuned ERNIE model is judged to be of practicality that meets the standard, and the trained fine-tuned ERNIE model is output as a car purchase intention recognition model; When K<K0, it is determined that the practicality of the trained fine-tuned ERNIE model does not meet the standard, and the data collection module is returned to obtain the natural language data of the second user, and the fine-tuned ERNIE model is trained according to the natural language data of the second user until the practicality of the trained fine-tuned ERNIE model meets the standard.

Images