Driver radar chart construction method and system based on multimodal user portrait

Abstract

The present invention relates to the field of information technology, and in particular to a method and system for constructing a driver radar chart based on multimodal user portraits, the method comprising: collecting structured data, unstructured text data and sequence data of drivers; preprocessing the structured data, unstructured text data and sequence data respectively, and extracting features from the preprocessed structured data, unstructured text data and sequence data respectively; splicing the extracted feature vectors, and using an attention mechanism to weightedly fuse the preliminary feature representations; determining multiple evaluation dimensions, and calculating the comprehensive scores under different evaluation dimensions by fusion feature calculation; and generating a driver radar chart based on the comprehensive scores. The present invention can generate a comprehensive and intuitive driver radar chart through the fusion and feature extraction of multimodal data, showing the comprehensive performance of drivers in multiple dimensions.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method and system for constructing a driver radar chart. Background Art

[0002] Existing driver evaluation and recommendation systems mostly rely on structured data, such as basic attributes of drivers (driving experience, number of completed orders, etc.) and transportation efficiency data. However, these data can only reflect the performance of drivers in specific dimensions and cannot fully evaluate the comprehensive capabilities of drivers. At the same time, unstructured text data (such as customer reviews, complaint records, etc.) and time series data (such as the time series of driver order completion, etc.) also contain a lot of information related to driver performance, but these data are usually ignored due to the complexity of processing. Summary of the invention

[0003] The purpose of the present invention is to provide a method for constructing a driver radar chart based on multimodal user portraits to solve the above technical problems;

[0004] The present invention also aims to provide a driver radar chart construction system based on multimodal user portraits to solve the above technical problems.

[0005] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0006] The driver radar chart construction method based on multimodal user portraits includes:

[0007] Step S1, collecting driver's structured data, unstructured text data and sequence data;

[0008] Step S2, preprocessing the structured data, the unstructured text data and the sequence data respectively, and extracting features from the preprocessed structured data, the unstructured text data and the sequence data respectively;

[0009] Step S3, concatenating the extracted feature vectors to obtain a preliminary feature representation, and using an attention mechanism to perform weighted fusion on the preliminary feature representation to obtain a fused feature representation;

[0010] Step S4, determining multiple evaluation dimensions, and calculating comprehensive scores under different evaluation dimensions by fusion feature calculation;

[0011] Step S5, generating a driver radar chart according to the comprehensive score.

[0012] Preferably, in step S1, the structured data includes driver attribute data and order data, the unstructured text data includes customer feedback and evaluation data and internal record data, and the sequence data is historical behavior data.

[0013] Preferably, in step S2, the method for preprocessing the structured data is to perform maximum and minimum normalization processing on the structured data, convert the structured data into a range of (0, 1), and the normalization expression is:

[0014]

[0015] Among them, x * represents the normalized result, x represents the original structured data, and x min Represents the minimum value of the structured data, x max Indicates the maximum value of the structured data.

[0016] Preferably, in step S2, the method for preprocessing the unstructured text data includes segmenting the unstructured text data and removing stop words.

[0017] Preferably, in step S2, the method for extracting features from the unstructured text data is:

[0018] The unstructured text data is vectorized using a pre-trained language model to obtain a high-dimensional feature representation of the unstructured text data, expressed as:

[0019] E text =BERT(T)

[0020] Where T represents the input text, BERT represents the language model, and E text Represents the extracted text feature vector.

[0021] Preferably, in step S2, the method for extracting features from the sequence data is:

[0022] The long short-term memory network is used to extract the temporal features in the sequence data, and the expression is:

[0023] h t =LSTM(x t ,h t-1 )

[0024] Among them, x t represents the current input in the sequence data, h t represents the current hidden state, h t-1 Indicates the hidden state at the previous moment.

[0025] Preferably, in step S3, the expression for weighted fusion of the preliminary feature representation using the attention mechanism is:

[0026]

[0027] Among them, F i represents the i-th type of modal feature, n represents the number of modal features, α i Represents the weight calculated by the attention mechanism.

[0028] Preferably, in step S4, the determined evaluation dimensions include at least a safe driving evaluation dimension, a transportation efficiency evaluation dimension, a service quality evaluation dimension and a work stability evaluation dimension.

[0029] Preferably, in step S4, the score of each evaluation dimension is calculated by fusion feature, and the expression is,

[0030]

[0031] Among them, w i represents the weight of the i-th feature, represents the value of the i-th feature on the j-th evaluation dimension, n represents the number of modal features, and score j Represents the score of the jth evaluation dimension.

[0032] The driver radar chart construction system based on multimodal user portraits includes:

[0033] A data collection module, used to collect structured data, unstructured text data, and sequence data;

[0034] A data preprocessing module, connected to the data collection module, for cleaning and preprocessing the structured data, the unstructured text data and the sequence data;

[0035] A feature extraction module, connected to the data preprocessing module, for extracting feature vectors from the structured data, the unstructured text data and the sequence data;

[0036] A feature fusion module, connected to the feature extraction module, is used to concatenate the feature vectors and perform weighted fusion through an attention mechanism;

[0037] A scoring calculation module, connected to the feature fusion module, is used to calculate the comprehensive scores under different evaluation dimensions by fusion feature calculation;

[0038] A radar chart generating module is connected to the score calculating module and is used to generate a driver radar chart according to the comprehensive score.

[0039] Beneficial effects of the present invention: Due to the adoption of the above technical solution, the present invention can generate a comprehensive and intuitive driver radar chart through the fusion and feature extraction of multimodal data, showing the comprehensive performance of the driver in multiple dimensions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of the steps of a method for constructing a driver radar map in an embodiment of the present invention;

[0041] Figure 2 A flowchart of a method for constructing a driver radar map according to an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of a driver radar diagram in an embodiment of the present invention;

[0043] Figure 4 The following is an architecture diagram of a driver radar chart construction system in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0046] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0047] Driver radar chart construction method based on multimodal user portraits, such as Figure 1 , Figure 2 As shown, including,

[0048] Step S1, collecting driver's structured data, unstructured text data and sequence data;

[0049] Step S2, preprocessing the structured data, the unstructured text data and the sequence data respectively, and extracting features from the preprocessed structured data, the unstructured text data and the sequence data respectively;

[0050] Step S3, concatenating the extracted feature vectors to obtain a preliminary feature representation, and using an attention mechanism to perform weighted fusion on the preliminary feature representation to obtain a fused feature representation;

[0051] Step S4, determining multiple evaluation dimensions, and calculating the comprehensive scores under different evaluation dimensions by fusion feature calculation;

[0052] Step S5, generating a driver radar chart based on the comprehensive score.

[0053] Specifically, the present invention processes driver index data, which includes structured data, unstructured text data, and sequence data of user behavior, to construct multimodal user portraits, and generates driver radar charts based on these portraits, thereby comprehensively displaying the comprehensive performance of drivers in different dimensions. The obtained driver radar chart can be used to evaluate the work performance of drivers and assist managers in task allocation and performance appraisal.

[0054] In a preferred embodiment, in step S1, the structured data includes driver attribute data and order data, the unstructured text data includes customer feedback and evaluation data and internal record data, and the sequence data is historical behavior data.

[0055] Specifically, the structured data includes driver attribute data and order data, among which:

[0056] Driver attribute data includes basic information about the driver, such as driving experience, license status, whether full-time, number of monthly orders, etc., which is used to reflect the driver's basic transportation capabilities and reliability.

[0057] Order data such as the completion time, loading and unloading locations, cargo type, transportation distance, etc. of each order reflects the driver's task execution efficiency and regional adaptability.

[0058] Unstructured text data includes customer feedback and evaluation data and internal record data, among which:

[0059] Customer feedback and evaluation data, such as customer comments, complaints, suggestions, etc., include subjective evaluations of the driver's service quality, which are used to reflect the driver's service attitude and overall experience.

[0060] Internal records, such as the company’s internal customer service records and driving behavior reports, reflect the driver’s compliance and service level.

[0061] Sequence data is the time series of the driver's past task execution, showing the driver's work frequency, order acceptance time distribution and task completion rate. Sequence data can reflect the driver's work habits and behavior patterns.

[0062] In a preferred embodiment, in step S2, the method for preprocessing the structured data is to perform maximum and minimum normalization processing on the structured data, converting the structured data into the range of (0, 1), and the normalization expression is:

[0063]

[0064] Among them, x * represents the normalized result, x represents the original structured data, and x min Represents the minimum value of structured data, x max Represents the maximum value of structured data.

[0065] Specifically, before preprocessing structured data, unstructured text data, and sequence data, it is necessary to clean the structured data, unstructured text data, and sequence data to remove outliers and incomplete data to ensure data quality and consistency;

[0066] Since the dimensions of structured data vary, normalization processing is required. For example, perform min-max normalization on structured data (such as the number of orders and task completion time) and convert it to the range of (0, 1). The formula for structured data processing is as follows: Standardize the structured data using the Min-Max normalization formula:

[0067]

[0068] where x * represents the normalization result, x represents the original structured data, and x min represents the minimum value of the structured data, and x max represents the maximum value of the structured data.

[0069] In a preferred embodiment, in step S2, the method for preprocessing unstructured text data includes tokenizing the unstructured text data and removing stop words.

[0070] Specifically, for unstructured text data, first perform preprocessing such as tokenization and stop word removal, and then use pre-trained models such as BERT for vectorization to obtain a high-dimensional feature representation of the text.

[0071] Tokenization refers to splitting a text into words or sub-word units (tokens). For example, the text "I am satisfied with the driver's service" is split into units such as "I", "driver", "service", "satisfied", etc.

[0072] Stop words in stop word removal refer to those words that appear frequently in the text but are not very helpful for semantic understanding, such as "of", "is", "in", "and", etc. Removing stop words is used to reduce computational complexity and improve the efficiency of the language model.

[0073] In a preferred embodiment, in step S2, the method for feature extraction of unstructured text data is

[0074] Use a pre-trained language model to vectorize the unstructured text data to obtain a high-dimensional feature representation of the unstructured text data. The expression is

[0075] E text = BERT(T)

[0076] Where T represents the input text, BERT represents the language model, and E text Represents the extracted text feature vector.

[0077] Specifically, the present invention extracts key information from customer reviews and internal records through the BERT model and converts it into a high-dimensional feature vector to reflect the driver's service quality and compliance. The BERT model can generate a context-related vector representation of a text or word by encoding text in a bidirectional manner. After each input text T is processed by the BERT model, a high-dimensional feature vector E is generated. text , the vector contains the semantic information of the text.

[0078] In a preferred embodiment, in step S2, the method for extracting features from sequence data is to use a long short-term memory network to extract temporal features from the sequence data, and the expression is:

[0079] h t =LSTM(x t ,h t-1 )

[0080] Among them, x t Represents the current input in the sequence data, h t represents the current hidden state, h t-1 Indicates the hidden state at the previous moment.

[0081] Specifically, the preprocessing method for sequence data in the present invention is to directly use the long short-term memory network (LSTM) to process the user behavior sequence data, extract the timing features in the behavior sequence, and capture the timing features in the driver's historical behavior through the LSTM model, which is used to analyze the driver's work habits and the stability of task completion.

[0082] LSTM can remember important information in long time series by continuously updating hidden states. In the formula,

[0083] h t =LSTM(x t ,h t-1 )

[0084] x t represents the input at the current moment, represents the element at time t in the historical behavior sequence, and h t is the output at the current moment, representing the hidden state at time t in the sequence, combining the contextual information of the current input and the hidden state at the previous moment, h t-1 It is the hidden state of the previous moment, which retains the historical dependency information.

[0085] The role of LSTM is to determine the retention and forgetting of information through gating mechanisms (such as forget gate, input gate, output gate, etc.), so as to effectively handle long-term dependencies and avoid gradient vanishing or exploding problems. Therefore, LSTM can extract meaningful features from the driver's historical behavior.

[0086] In a preferred embodiment, in step S3, the expression for weighted fusion of the preliminary feature representation using the attention mechanism is:

[0087]

[0088] Among them, F i represents the i-th type of modal feature, n represents the number of modal features, α i Represents the weight calculated by the attention mechanism.

[0089] Specifically, when performing multimodal feature fusion, data of different modalities (structured data, unstructured text data, and sequence data) are first concatenated to obtain a preliminary feature representation:

[0090] F concat =[F struct ,F text ,F seq ]

[0091] Among them, F struct The feature vector representing the structured data, F text The feature vector representing the unstructured text data, F seq A feature vector representing the sequence data.

[0092] In order to further optimize the weights of different modal features, the attention mechanism is used to perform weighted fusion on the concatenated features. The expression is:

[0093]

[0094] Among them, F represents the i-th type of modal feature, n represents the number of modal features, α i Represents the weight calculated by the attention mechanism.

[0095] In a preferred embodiment, in step S4, the determined evaluation dimensions include at least a safe driving evaluation dimension, a transportation efficiency evaluation dimension, a service quality evaluation dimension, and a work stability evaluation dimension.

[0096] Specifically, the dimension determination in the present invention includes:

[0097] The safe driving assessment dimension evaluates the driver's safe driving level based on the driver's historical violation records, driving experience and other safety data.

[0098] The transportation efficiency evaluation dimension evaluates the driver's transportation speed and accuracy based on order completion time, task execution efficiency, etc.

[0099] The service quality evaluation dimension evaluates the driver’s service attitude and customer satisfaction through unstructured data such as customer reviews and complaint records.

[0100] The job stability assessment dimension evaluates the driver's job stability and activity based on sequence data such as the driver's task execution frequency and order acceptance interval.

[0101] In a preferred embodiment, in step S4, the score of each evaluation dimension is calculated by fusion features, and the expression is:

[0102]

[0103] Among them, w i represents the weight of the i-th feature, represents the value of the i-th feature on the j-th evaluation dimension, n represents the number of modal features, and score j Represents the score of the jth evaluation dimension.

[0104] Comprehensive score calculation: The score of each dimension is calculated by fusion features, and the formula is:

[0105]

[0106] Among them, w i represents the weight of the i-th feature, represents the value of the i-th feature on the j-th evaluation dimension, n represents the number of modal features, and score j Represents the score of the jth evaluation dimension.

[0107] In the driver evaluation dimension, taking the safe driving evaluation dimension as an example, assuming that the two features corresponding to the safe driving evaluation dimension are the number of violations and driving experience, and the weight of the number of violations is 0.7 and the weight of driving experience is 0.3, then the comprehensive score of the driver in the safe driving evaluation dimension is calculated based on the eigenvalues ​​of the two features and the corresponding weights.

[0108] The other evaluation dimensions are calculated in the same way, and finally the comprehensive scores of different evaluation dimensions are obtained. According to the comprehensive scores of each evaluation dimension, a driver radar chart is generated. The radar chart is as follows: Figure 3 As shown in the figure, each axis represents a dimension, and the higher the score, the better the driver's performance in that dimension. The radar chart intuitively displays the driver's comprehensive ability through the performance of each dimension.

[0109] Subsequent application scenarios of radar charts include

[0110] Performance evaluation: Radar charts can be used for daily performance evaluation of drivers, helping company management to quickly understand the overall performance of drivers.

[0111] Task allocation: Based on the performance in the radar chart, tasks are reasonably allocated to drivers with outstanding performance to improve the efficiency of task completion.

[0112] Incentive mechanism: The radar chart can serve as an important basis for driver rewards and promotions, motivating drivers to improve their service quality.

[0113] Driver radar chart construction system based on multimodal user portraits, such as Figure 4 As shown, including,

[0114] Data collection module 1, used to collect structured data, unstructured text data and sequence data;

[0115] The data preprocessing module 2 is connected to the data collection module 1 and is used to clean and preprocess the structured data, unstructured text data and sequence data;

[0116] A feature extraction module 3, connected to the data preprocessing module 2, is used to extract feature vectors from structured data, unstructured text data and sequence data;

[0117] Feature fusion module 4, connected to feature extraction module 3, is used to concatenate feature vectors and perform weighted fusion through attention mechanism;

[0118] The score calculation module 5 is connected to the feature fusion module 4 and is used to calculate the comprehensive scores under different evaluation dimensions by fusion feature calculation;

[0119] The radar chart generating module 6 is connected to the score calculating module 4 and is used to generate a driver radar chart according to the comprehensive score.

[0120] In summary, through the fusion and feature extraction of multimodal data, the present invention can generate a comprehensive and intuitive driver radar chart to show the comprehensive performance of drivers in multiple dimensions. Compared with traditional evaluation methods, the present invention can more accurately reflect the overall ability of drivers and effectively assist managers in making scientific decisions in task allocation, performance evaluation and other aspects.

[0121] The above description is only a preferred embodiment of the present invention, and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A driver radar chart construction method based on multimodal user portraits, characterized in that: include, Step S1, collecting driver's structured data, unstructured text data and sequence data; Step S2, preprocessing the structured data, the unstructured text data and the sequence data respectively, and extracting features from the preprocessed structured data, the unstructured text data and the sequence data respectively; Step S3, concatenating the extracted feature vectors to obtain a preliminary feature representation, and using an attention mechanism to perform weighted fusion on the preliminary feature representation to obtain a fused feature representation; Step S4, determining multiple evaluation dimensions, and calculating comprehensive scores under different evaluation dimensions by fusion feature calculation; Step S5, generating a driver radar chart according to the comprehensive score.

2. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S1, the structured data includes driver attribute data and order data, the unstructured text data includes customer feedback and evaluation data and internal record data, and the sequence data is historical behavior data.

3. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S2, the method for preprocessing the structured data is to perform maximum and minimum normalization processing on the structured data, convert the structured data into the range of (0, 1), and the normalization expression is: Among them, x * represents the normalized result, x represents the original structured data, and x min represents the minimum value of the structured data, x max Indicates the maximum value of the structured data.

4. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S2, the method for preprocessing the unstructured text data includes segmenting the unstructured text data and removing stop words.

5. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S2, the method for extracting features from the unstructured text data is: The unstructured text data is vectorized using a pre-trained language model to obtain a high-dimensional feature representation of the unstructured text data, expressed as: E text =BERT(T) Where T represents the input text, BERT represents the language model, and E text Represents the extracted text feature vector.

6. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S2, the method for extracting features from the sequence data is: The long short-term memory network is used to extract the temporal features in the sequence data, and the expression is: h t =LSTM(x t ,h t-1 ) Among them, x t represents the current input in the sequence data, h t represents the current hidden state, h t-1 Represents the hidden state at the previous moment.

7. The method for constructing a driver radar chart based on multimodal user portrait according to claim 1, characterized in that: In step S3, the expression for weighted fusion of the preliminary feature representation using the attention mechanism is: Among them, F i represents the i-th type of modal feature, n represents the number of modal features, α i Represents the weight calculated by the attention mechanism.

8. The method for constructing a driver radar chart based on multimodal user portraits according to claim 1, characterized in that: In step S4, the determined evaluation dimensions include at least a safe driving evaluation dimension, a transportation efficiency evaluation dimension, a service quality evaluation dimension, and a work stability evaluation dimension.

9. The method for constructing a driver radar chart based on multimodal user portraits according to claim 1, characterized in that: In step S4, the score of each evaluation dimension is calculated by fusion feature, and the expression is: Among them, w i represents the weight of the i-th feature, represents the value of the i-th feature on the j-th evaluation dimension, n represents the number of modal features, and score j Represents the score of the jth evaluation dimension.

10. A driver radar chart construction system based on multimodal user portraits, characterized in that: include, A data collection module (1), used for collecting structured data, unstructured text data and sequence data; A data preprocessing module (2), connected to the data collection module (1), for cleaning and preprocessing the structured data, the unstructured text data and the sequence data; A feature extraction module (3), connected to the data preprocessing module (2), for extracting feature vectors from the structured data, the unstructured text data and the sequence data; A feature fusion module (4), connected to the feature extraction module (3), is used to concatenate the feature vectors and perform weighted fusion through an attention mechanism; A score calculation module (5), connected to the feature fusion module (4), is used to calculate the comprehensive scores under different evaluation dimensions by fusion feature calculation; A radar chart generating module (6) is connected to the score calculating module (5) and is used to generate a driver radar chart according to the comprehensive score.

Images