A method and device of a deep learning-based online car-hailing driver and passenger mutual evaluation system

Abstract

The application relates to a method and device for a deep learning-based online car-hailing driver and passenger mutual evaluation system, and in particular to the field of artificial intelligence. Basic information of the driver and passenger and mutual evaluation data between the driver and passenger are collected and preprocessed, the statistical conversion is replaced by a standardized score number and scaling processing is performed, which helps to eliminate the dimensional influence and deviation between the score numbers and map the score results to a more suitable range, so that the score is more comparable. Emotion tendency features and keyword frequency features are extracted and expressed, which helps to understand the key content and expression mode of the text and better understand the communication and evaluation between the driver and passenger. A word embedding model is used to convert each word into a corresponding word vector and perform weighted average to obtain an overall vector representation. An LSTM unit is selected to build a recurrent neural network model, and a sum score of the driver and passenger is output through a full connection layer, which can comprehensively consider the emotion tendency and evaluation of the driver and passenger.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically, to a method and apparatus for a deep learning-based ride-hailing driver-passenger mutual evaluation system. Background Technology

[0002] With the rapid development of the mobile internet, ride-hailing services have become an indispensable part of people's lives. Since drivers and passengers have different values ​​and lifestyles, establishing a fair and just mutual evaluation system in this diversified context has become an important aspect of improving the quality of ride-hailing services.

[0003] In traditional driver-passenger peer review systems, drivers can obtain passenger ratings, but passengers cannot obtain genuine feedback from drivers, creating an information asymmetry. Traditional star ratings often fail to accurately reflect user satisfaction and are easily influenced by a single unpleasant experience. Furthermore, existing driver-passenger peer review systems lack effective ways to resolve conflicts between drivers and passengers, resulting in problems not being properly addressed and impacting the driver-passenger experience. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a method and apparatus for a ride-hailing driver-passenger mutual evaluation system based on deep learning, thereby resolving the issues raised in the background section.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for a ride-hailing driver-passenger mutual evaluation system based on deep learning, comprising the following steps:

[0006] S101. Collect and preprocess the basic information of drivers and passengers themselves and the mutual evaluation data between drivers and passengers, establish an initial vocabulary and merge and update it, use the updated unified vocabulary to perform replacement statistics to convert it into standardized score numbers, and scale the standardized score numbers.

[0007] S102. Obtain the basic information of the driver and passenger and the number of positive and negative words in the mutual evaluation data between the driver and passenger, and extract the sentiment tendency features and represent the sentiment polarity based on the comparison of word counts. Statistically count the number of keywords and characters in the basic information of the driver and passenger and the mutual evaluation data between the driver and passenger, and extract the keyword frequency features. Use the word embedding model to map the words to a continuous vector space, convert each word into a corresponding word vector, and perform a weighted average to obtain the overall vector representation.

[0008] S103. Output the sentiment classification labels and corresponding sentiment analysis scores of the features. The driver and passenger and passenger and driver ...

[0009] S104. The recurrent neural network model is integrated into the driver-rider mutual evaluation system through server deployment and a data transmission interface is set up, and the corresponding suggestions and matching measures are obtained through the real-time sum score of the driver and the rider;

[0010] In a preferred embodiment, in S101, the basic information of the driver and the rider and the mutual evaluation data between the driver and the rider are collected, the missing values of the basic information of the driver and the rider and the mutual evaluation data between the driver and the rider are detected, the records corresponding to the basic information and the mutual evaluation data of the missing values are selected to be deleted, the repeated records of the basic information of the driver and the rider and the mutual evaluation data between the driver and the rider are detected, and the repeated records are selected to be deleted, the basic information of the driver and the rider and the mutual evaluation data between the driver and the rider are subjected to text processing, an initial vocabulary table is established and synonyms and near-synonyms are found through WordNet, for English text, different word forms are merged into the same root by using stem extraction, for Chinese text, the text is segmented by using a segmentation tool, and the synonyms and near-synonyms in the segmentation results are merged, the vocabulary table is updated, the updated unified vocabulary table is used for replacement, the original score number range is determined, and the standardized score number is converted and counted, and the specific formula is:

[0011]

[0012] wherein, Score S represents the standardized score number, Score O represents the original score number, Mean represents the average value of the original score number, Deviation S represents the standard deviation value of the original score number, the new score number range is determined, and the standardized score number is subjected to scaling processing, and the specific formula is:

[0013] Score d = Score S * Range n + Min n

[0014] wherein, Score d represents the conversion of the standardized score number to the score number in the new score range, Range n represents the length of the new score range, Min n represents the minimum value of the scaled score number range.

[0015] In a preferred embodiment, in the S102, the basic information of the driver and the mutual evaluation data between the drivers are traversed by using the sentiment dictionary, and the number of positive and negative words contained in the basic information of the driver and the mutual evaluation data between the drivers is obtained; the sentiment tendency feature is extracted and the sentiment polarity is represented according to the word quantity comparison; the keyword frequency feature is extracted by counting the number of keywords and characters of the basic information of the driver and the mutual evaluation data between the drivers; a word embedding model is selected and inputted into the basic information of the driver and the mutual evaluation data between the drivers for training, wherein the word embedding model maps words into a continuous vector space, converts each word into a corresponding word vector, and obtains an overall vector representation by weighted average.

[0016] In a preferred embodiment, in the S103, the driver and passenger dimensions are divided, and the StructBERT model is selected; the sentiment tendency feature and the keyword frequency feature are inputted, and the sentiment polarity of the features is returned through the sentiment dictionary, the keywords and the number of characters; the StructBERT model is loaded by using the Transformers library and is set to the evaluation mode; the input is converted into an encoding sequence by using the token i zer statement and is passed to the StructBERT model; the sentiment classification label and the corresponding sentiment analysis score of the features are outputted; the comprehensive score is obtained by weighted average according to the sentiment classification label and the corresponding sentiment analysis score of the driver and passenger dimensions, and the specific formula is as follows:

[0017]

[0018] wherein, C z represents the comprehensive score, C sc represents the driver sentiment analysis score, W s represents the driver weight, C cs represents the passenger sentiment analysis score, W c represents the passenger weight.

[0019] Further, the LSTM unit is selected to build a recurrent neural network model, and the accumulation speed of the features is controlled by introducing a gating mechanism, wherein the gating mechanism includes an input gate, a forget gate and an output gate; the input gate is divided into a sentiment score input layer and a basic information input layer, and the basic information of the driver, the mutual evaluation data between the drivers and the comprehensive score at the current time step and the hidden state at the previous time step are inputted; the activation value of the input gate is calculated by using the Sigmoid activation function and is converted into a value between 0 and 1, and the specific formula is as follows:

[0020] R l = α(W l *[I t-1 ,x t ]+b l )

[0021] wherein R l represents the activation value of the input gate, a() represents the Sigmoid activation function, W l represents the weight matrix, I t-1 represents the hidden state of the previous time step, x t represents the basic information of the driver and the passenger, the mutual evaluation data between the driver and the passenger, and the comprehensive score input at the current time step, b l represents the bias term, the output gate selects the fully connected layer and calculates the activation value of the output gate through the Sigmoid activation function according to the basic information of the driver and the passenger, the mutual evaluation data between the driver and the passenger, and the comprehensive score input at the current time step, the hidden state of the previous time step, and the cell state at the current time step, and converts it to a value between 0 and 1, and the specific formula is:

[0022] R o = a(W o *[I t-1 , x t ] + b o )

[0023] wherein R o represents the activation value of the forgetting gate, a() represents the Sigmoid activation function, W o represents the weight matrix, I t-1 represents the hidden state of the previous time step, x t represents the basic information of the driver and the passenger, the mutual evaluation data between the driver and the passenger, and the comprehensive score input at the current time step, b o represents the bias term, and the sum score of the driver and the passenger is output through the fully connected layer.

[0024] In a preferred embodiment, in S104, the recurrent neural network model is integrated into the driver and passenger mutual evaluation system through server deployment and a data transmission interface is set, / api / feedback is used as the URL endpoint of the interface, JSON is defined as the transmission format, the corresponding suggestions and matching measures are obtained through the real-time sum score of the driver and the passenger, a user interface is provided to allow the user to view the evaluation results, suggestions and matching measures, and a feedback function is provided, a monitoring mechanism is set to monitor the evaluation data and the execution of the suggestions in real time.

[0025] The application also provides a device of a ride-hailing driver and passenger mutual evaluation system based on deep learning, comprising a driver and passenger mutual evaluation platform, a vocabulary table, a StructBERT model and a Beidou navigation API, wherein the driver and passenger mutual evaluation platform is used to collect basic information of the driver and passenger and mutual evaluation data between the driver and passenger, the vocabulary table is used to find synonyms and near-synonyms through WordNet, the StructBERT model is used to obtain the comprehensive score given by one party to the other party in an order, and the Beidou navigation API is used to provide a better route selection.

[0026] The application has the advantages that: the statistical conversion into a standardized score number and the scaling processing help to eliminate the dimensional influence and deviation between score numbers and map the score results to a more suitable range, so that the score is more comparable, the sentiment color in the text is quantified through the sentiment analysis feature extraction, which helps to understand the emotional attitude and tendency between the driver and passenger, the keyword frequency feature extraction helps to understand the key content and expression method of the text, and the driver and passenger communication and evaluation are better understood, the word embedding model helps to convert the text data into a numerical representation form that can be processed by the machine learning model, the sentiment tendency and evaluation of the driver and passenger are comprehensively considered through the StructBERT model combined with the gated recurrent neural network model, the server deployment and user interface are used to generate suggestions and matching measures in real time, which helps the driver and passenger to improve the service quality and experience, and the driver and passenger can conveniently feedback the evaluation results and suggestions. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The method flowchart of the application is shown in the figure; DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0029] In the description of the application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0030] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in this application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for purposes of explanation, specific details are set forth. It will be apparent to those skilled in the art that the present application can be practiced without the specific details. In other instances, well-known structures and processes are not shown in detail to avoid obscuring the application. Thus, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features presented herein.

[0031] The present embodiment provides a method and device for a deep learning-based online car-hailing driver and passenger mutual evaluation system, specifically comprising the following steps: Figure 1 The present embodiment provides a method and device for a deep learning-based online car-hailing driver and passenger mutual evaluation system, specifically comprising the following steps:

[0032] S101. Collect and pre-process the basic information of the driver and passenger themselves and the mutual evaluation data between the driver and passenger, establish an initial vocabulary table and merge and update it, use the updated unified vocabulary table for replacement, and convert it into standardized score numbers, and scale the standard score numbers;

[0033] Further, collect the basic information of the driver and passenger themselves and the mutual evaluation data between the driver and passenger, including name, gender, age, contact information, evaluation content, score, and timestamp, detect the missing values of the basic information of the driver and passenger themselves and the mutual evaluation data between the driver and passenger, delete the records of the corresponding basic information and mutual evaluation data for the missing values, detect the repeated records of the basic information of the driver and passenger themselves and the mutual evaluation data between the driver and passenger, and delete the repeated records, perform text processing on the basic information of the driver and passenger themselves and the mutual evaluation data between the driver and passenger, for removing unnecessary punctuation marks and special characters, and only retaining the text and score number content, establish an initial vocabulary table and find synonyms and near-synonyms through WordNet, for English text, use stem extraction to merge different word forms into the same root, for Chinese text, use a word segmentation tool to segment the text, and merge synonyms and near-synonyms in the segmentation results, update the vocabulary table to ensure that all merged words are included in the vocabulary table, use the updated unified vocabulary table for replacement, determine the range of original score numbers, and convert them into standardized score numbers, the specific formula is:

[0034]

[0035] wherein Score S represents the standardized score number, Score OThe original score number, Mean represents the average of the original score number, Deviation S The standard deviation value of the original score number, determine the new score number range, and scale the standard score number, the specific formula is:

[0036] Score d = Score S *Range n +Min n

[0037] Wherein, Score d The normalized score number is converted into a score number in the new score range, Range n The length of the new score range, Min n The minimum value of the scaled score number range.

[0038] S102. Obtain the basic information of the driver and the passenger, and the number of positive and negative words contained in the mutual evaluation data between the driver and the passenger, and extract the emotional tendency feature and represent the emotional polarity according to the word quantity comparison, count the keywords and character numbers of the basic information of the driver and the passenger and the mutual evaluation data between the driver and the passenger, and extract the keyword frequency feature, map the words to a continuous vector space using a word embedding model, convert each word to a corresponding word vector and obtain the overall vector representation by weighted average;

[0039] Further, the basic information of the driver and the passenger and the mutual evaluation data between the driver and the passenger are traversed using the sentiment dictionary, and the number of positive and negative words contained in the basic information of the driver and the passenger and the mutual evaluation data between the driver and the passenger is obtained, the emotional tendency feature is extracted and the emotional polarity is represented according to the word quantity comparison, when the number of positive words is more than the number of negative words, the emotional polarity is determined to be positive, when the number of positive words is less than the number of negative words, the emotional polarity is determined to be negative, and when the number of positive words is equal to the number of negative words, the emotional polarity is determined to be neutral, the keywords and character numbers of the basic information of the driver and the passenger and the mutual evaluation data between the driver and the passenger are counted and the keyword frequency feature is extracted, a word embedding model is selected and input to the basic information of the driver and the passenger and the mutual evaluation data between the driver and the passenger for training, wherein the word embedding model maps words to a continuous vector space, which is used to have similar words in the vector space. Close distance, convert each word to a corresponding word vector and obtain the overall vector representation by weighted average.

[0040] S103. Output the emotional classification label and the corresponding emotional analysis score of the feature, and obtain the comprehensive score by weighted average of the driver and the passenger in two dimensions according to the emotional classification label and the corresponding emotional analysis score, select the LSTM unit to build a recurrent neural network model and output the total score of the driver and the passenger through the full connection layer;

[0041] Further, the driver and passenger dimensions are divided and the StructBERT model is selected to obtain the comprehensive score given by the driver and passenger to each other in an order, the sentiment tendency features and keyword frequency features are input, and the sentiment polarity of the features is returned through the sentiment dictionary, keywords and character number. The StructBERT model is loaded by using the Transformers library and set to evaluation mode. The input is converted into an encoding sequence by using the token i zer statement and passed to the StructBERT model. The sentiment classification label and the corresponding sentiment analysis score of the features are output. The comprehensive score is obtained by weighted average of the driver and passenger dimensions according to the sentiment classification label and the corresponding sentiment analysis score. The specific formula is:

[0042]

[0043] wherein C z represents the comprehensive score, C sc represents the driver sentiment analysis score, W s represents the driver weight, C cs represents the passenger sentiment analysis score, W c represents the passenger weight.

[0044] Further, the LSTM unit is selected to build a recurrent neural network model. The accumulation speed of the features is controlled by introducing a gating mechanism, including selectively adding new features and selectively forgetting previously accumulated features. The gating mechanism includes an input gate, a forget gate and an output gate. The input gate is divided into a sentiment score input layer and a basic information input layer. The basic information of the driver and passenger, the mutual evaluation data between the driver and passenger, and the comprehensive score of the current time step input are input. An LSTM layer is added to the output of the first layer LSTM input gate to capture time series in a deeper layer. The activation value of the input gate is calculated by using a Sigmoid activation function and converted into a value between 0 and 1. The specific formula is:

[0045] R l = α(W l *[I t-1 ,x t ]+b l )

[0046] wherein R l represents the activation value of the input gate, α() represents the Sigmoid activation function, W l represents the weight matrix, I t-1 represents the hidden state of the previous time step, x t represents the basic information of the driver and passenger, the mutual evaluation data between the driver and passenger, and the comprehensive score input at the current time step, and b lThe bias term, the output gate selection fully connected layer and the basic information of the driver and the passenger input at the current time step, mutual evaluation data between the driver and the passenger and a comprehensive score, the hidden state at the last time step and the cell state at the current time step are output, the activation value of the output gate is calculated through a Sigmoid activation function and converted into a value between 0 and 1, and the specific formula is:

[0047] R o = alpha(W o *[I t-1 ,x t ]+b o )

[0048] Wherein, R o represents the activation value of the forgetting gate, alpha() represents the Sigmoid activation function, W o represents the weight matrix, I t-1 represents the hidden state at the last time step, x t represents the basic information of the driver and the passenger, mutual evaluation data between the driver and the passenger and a comprehensive score input at the current time step, b o represents the bias term, and the sum score of the driver and the passenger is output through a fully connected layer.

[0049] S104. The recurrent neural network model is integrated into the driver and passenger mutual evaluation system through server deployment and a data transmission interface is set, and corresponding suggestions and matching measures are obtained through the real-time sum score of the driver and the passenger.

[0050] Further, the recurrent neural network model is integrated into the driver and passenger mutual evaluation system through server deployment and a data transmission interface is set, the / api / feedback is used as the URL endpoint of the interface, the JSON is defined as the transmission format, the corresponding suggestions and matching measures are obtained through the real-time sum score of the driver and the passenger, when multiple passengers reflect that the service attitude of the driver is not good, a training plan for improving the service attitude of the driver is proposed, when the passenger reflects that the navigation route is not accurate, the Beidou navigation API is connected to provide a better route selection, a user interface is provided to allow the user to view the evaluation results, suggestions and matching measures, and a feedback function is provided, a monitoring mechanism is set to monitor the evaluation data and the suggestion execution in real time.

[0051] An embodiment of the present application also provides a device of a driver and passenger mutual evaluation system based on deep learning, which comprises a driver and passenger mutual evaluation platform, a vocabulary table, a StructBERT model and a Beidou navigation API, wherein the driver and passenger mutual evaluation platform is used for collecting the basic information of the driver and the passenger and mutual evaluation data between the driver and the passenger, the vocabulary table is used for finding synonyms and near synonyms through WordNet, the StructBERT model is used for obtaining the comprehensive score given by the driver and the passenger to each other in an order, and the Beidou navigation API is used for providing a better route selection.

[0052] It should be noted that the descriptions of the various embodiments are each given with emphasis on certain features of the embodiments. The descriptions of the various embodiments are not meant to be taken in a literal sense, and the features of the various embodiments can be combined with each other.

[0053] Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0054] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.

[0055] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.

[0057] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0058] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.

Claims

1. A method for a deep learning-based ride-hailing driver-passenger peer review system, characterized in that, Specifically, the following steps are included: S101. Collect and preprocess the basic information of drivers and passengers and the mutual evaluation data between drivers and passengers, establish an initial vocabulary list and merge and update it, replace it with the updated unified vocabulary list, statistically convert it into standardized score numbers, and scale the standardized score numbers. S102. Obtain the basic information of the driver and passenger and the number of positive and negative words in the mutual evaluation data between the driver and passenger, and extract the sentiment tendency features and represent the sentiment polarity based on the comparison of word counts. Statistically count the number of keywords and characters in the basic information of the driver and passenger and the mutual evaluation data between the driver and passenger, and extract the keyword frequency features. Use the word embedding model to map the words to a continuous vector space, convert each word into a corresponding word vector, and perform a weighted average to obtain the overall vector representation. S103. Select the StructBERT model, input sentiment tendency features and keyword frequency features, output the sentiment classification labels of the features and the corresponding sentiment analysis scores, calculate the weighted average of the driver and passenger dimensions and the passenger and driver based on the sentiment classification labels and the corresponding sentiment analysis scores to obtain the comprehensive score, select LSTM units to build a recurrent neural network model and output the sum score of the driver and passenger through a fully connected layer; S104. Deploy and integrate the recurrent neural network model into the driver-passenger mutual evaluation system via a server and set up a data transmission interface. Obtain corresponding suggestions and matching measures based on the real-time sum of scores from both the driver and passenger.

2. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 1, characterized in that: In step S101, the specific preprocessing steps are as follows: detecting missing values ​​in the driver's and passenger's basic information and the mutual evaluation data between the driver and passenger; deleting the corresponding basic information and mutual evaluation data records for missing values; detecting duplicate records in the driver's and passenger's basic information and the mutual evaluation data between the driver and passenger, and deleting duplicate records; and performing text processing on the driver's and passenger's basic information and the mutual evaluation data between the driver and passenger.

3. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 2, characterized in that: The specific steps of the text processing are as follows: establish an initial vocabulary and find synonyms and near-synonyms through WordNet, update the vocabulary, replace it with the updated unified vocabulary, determine the range of the original score numbers, convert them into standardized score numbers, determine the new range of score numbers, and scale the standardized score numbers.

4. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 3, characterized in that: The specific formula for converting the statistics into standardized score numbers is as follows: ； in, Represents standardized score numbers. This represents the original score number. This represents the average of the original score numbers. The scaling formula represents the standard deviation of the original score numbers. ； in, This means converting standardized rating numbers into rating numbers within a new rating range. Indicates the length of the new rating range. This represents the minimum value within the range of scaled rating numbers.

5. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 1, characterized in that: In step S102, the basic information of the driver and passenger and the number of positive and negative words in the mutual evaluation data between the driver and passenger are obtained. Based on the comparison of word counts, sentiment tendency features are extracted and sentiment polarity is represented. The number of keywords and characters in the basic information of the driver and passenger and the mutual evaluation data between the driver and passenger are counted and keyword frequency features are extracted. A word embedding model is selected and the basic information of the driver and passenger and the mutual evaluation data between the driver and passenger are input for training. Each word is converted into a corresponding word vector and a weighted average is performed to obtain the overall vector representation.

6. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 1, characterized in that: In step S103, the driver and passenger dimensions are divided into two dimensions: driver and passenger. The sentiment polarity of the features is returned through the sentiment dictionary, keywords, and character count. The StructBERT model is loaded using the Transformers library and set to evaluation mode. The tokenizer statement is used to convert the input into an encoded sequence and pass it to the StructBERT model. The sentiment classification label of the feature and the corresponding sentiment analysis score are output. The driver and passenger dimensions and the passenger and passenger dimensions are weighted and averaged according to the sentiment classification label and the corresponding sentiment analysis score to obtain the comprehensive score.

7. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 6, characterized in that: The specific formula for obtaining the comprehensive score by weighted average is as follows: ； in, This represents the overall score. This indicates the driver and passenger sentiment analysis score. Indicates driver weight, This indicates the passenger's sentiment analysis score. This indicates the passenger weight.

8. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 1, characterized in that: In step S103, the specific steps for selecting an LSTM unit to build a recurrent neural network model and outputting the sum score of the driver and passenger through a fully connected layer are as follows: By introducing a gating mechanism to control the accumulation speed of features, the input gate is divided into an emotion score input layer and a basic information input layer. Based on the basic information of the driver and passenger themselves, the mutual evaluation data between the driver and passenger, the comprehensive score, and the hidden state of the previous time step, the activation value of the input gate is calculated using the Sigmoid activation function and converted into a value between 0 and 1. The specific formula is as follows: ； in, This represents the activation value of the input gate. This represents the Sigmoid activation function. Represents the weight matrix. This indicates the hidden state of the previous time step. This indicates the basic information of the driver and passenger, the mutual evaluation data between the driver and passenger, and the overall score input at the current time step. The bias term is represented by a fully connected layer for the output gate. Based on the driver and passenger's basic information, mutual evaluation data, and overall score input at the current time step, the hidden state of the previous time step, and the cell state of the current time step, the activation value of the output gate is calculated using the Sigmoid activation function and converted into a value between 0 and 1. The specific formula is as follows: ； in, This represents the activation value of the forget gate. This represents the Sigmoid activation function. Represents the weight matrix. This indicates the hidden state of the previous time step. This indicates that the driver and passenger's basic information, mutual evaluation data, and overall score are input at the current time step. This indicates the bias term.

9. The method for a deep learning-based ride-hailing driver-passenger mutual evaluation system according to claim 1, characterized in that: In step S104, the recurrent neural network model is deployed and integrated into the driver-passenger mutual evaluation system through the server and a data transmission interface is set up. The URL endpoint of the interface is used as / api / feedback. The corresponding suggestions and matching measures are obtained through the real-time sum of scores between the driver and the passenger. A user interface and feedback function are provided, and a monitoring mechanism is set up to monitor the evaluation data and the implementation of suggestions in real time.

10. A device for a deep learning-based ride-hailing driver-passenger peer review system, characterized in that, The device includes: A method for implementing a deep learning-based ride-hailing driver-passenger mutual evaluation system according to any one of claims 1-9, comprising a driver-passenger mutual evaluation platform, a vocabulary, a StructBERT model, and instructions from the BeiDou Navigation API.

Images