Traffic complaint processing method, device and related equipment based on text mining

Abstract

This application provides a method, apparatus, and related equipment for handling traffic complaints based on text mining. For traffic complaint texts to be mined, this application determines the target recognition segments of the text through a debugged text keyword recognition network, and instructs the text keyword recognition network to determine whether each target recognition segment includes a set type of traffic-related entity. Then, by using the entity description vector of the traffic-related entity corresponding to each target recognition segment and the corresponding target inference result, it can be determined whether the traffic-related entity corresponding to each target recognition segment is a set type of traffic-related entity. This ensures the accuracy of identifying the set type of traffic-related entities, overcomes the problem of traffic complaint information identification errors, and achieves efficient, accurate, and batch extraction of complaint information, providing convenience for traffic management work.

Description

Technical Field

[0001] The present application relates to but is not limited to the fields of text data processing and machine learning technology, and in particular to a traffic complaint handling method, apparatus and related equipment based on text mining. Background Art

[0002] With the continuous advancement of urbanization, traffic congestion has become a major challenge in people's daily lives. At the same time, the popularity of the Internet and social media has also provided people with a convenient channel to express complaints and dissatisfaction with traffic problems. However, handling a large amount of traffic complaint information is a huge challenge for management departments and related companies. In order to better handle these traffic complaint information, natural language processing and text mining technologies are widely used. Based on this, relevant information can be effectively extracted from massive text data, speeding up problem solving and improving public satisfaction. Specifically, text mining technology can help management departments and companies quickly and accurately understand key information such as the main issues of complaints, the sources of complaints, and the frequency of complaints. Among them, how to ensure the accuracy of the mined text entities is a technical issue of concern in this field. Summary of the invention

[0003] In view of this, the embodiments of the present application at least provide a traffic complaint handling method, apparatus and related equipment based on text mining.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides a traffic complaint handling method based on text mining, which is applied to a traffic complaint handling device. The method includes:

[0006] Determine an initial text description vector of a traffic complaint text to be mined, input the initial text description vector of the traffic complaint text to be mined into a text keyword recognition network, and obtain one or more target recognition paragraphs and their target reasoning results; wherein each of the reasoning results is used to indicate whether the target recognition paragraph corresponding to the reasoning of the text keyword recognition network includes a set type of traffic-related entity;

[0007] Determining whether the target recognition section includes a set type of traffic-related entity through the target reasoning result of each target recognition section;

[0008] The text keyword recognition network includes the following operations during debugging:

[0009] Inputting the initial text description vector of each traffic complaint text sample into the basic text network, obtaining target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample, each of which indicates whether the target recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity;

[0010] The total debugging error is determined by the target reasoning results and actual entity information of each target recognition paragraph, the basic text network is debugged by each traffic complaint text sample until the total debugging error meets the set requirements, and the debugging is stopped, and the text network when the debugging is stopped is determined as a text keyword recognition network, and each actual entity information is used to indicate whether the corresponding target recognition paragraph includes a set type of traffic-related entity; wherein the process of determining each target reasoning result by the basic text network includes:

[0011] Through the initial text description vector of each traffic complaint text sample, one or more initial recognition paragraphs of the traffic complaint text sample are repeatedly adjusted for a set number of rounds to obtain the corresponding target recognition paragraph, wherein each round of optimization includes: adjusting each first recognition paragraph through the initial text description vector of the traffic complaint text sample to obtain a second recognition paragraph and a corresponding initial reasoning result, each initial reasoning result indicates whether the second recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity, the first recognition paragraph corresponding to each round of optimization is the second recognition paragraph obtained by the previous round of adjustment, the first recognition paragraph corresponding to the first adjustment is each of the initial recognition paragraphs, and the second recognition paragraph obtained by the last round of adjustment is the target recognition paragraph of the traffic complaint text sample;

[0012] The target reasoning results of each target recognition paragraph of the traffic complaint text sample are determined by the initial reasoning results of the second recognition paragraph obtained through each round of optimization.

[0013] In some embodiments, for each of the traffic complaint text samples, determining one or more initial recognition paragraphs of the traffic complaint text sample by using the initial text description vector of the traffic complaint text sample includes:

[0014] Determine each candidate identification paragraph of the traffic complaint text sample by using the initial text description vector of the traffic complaint text sample;

[0015] For each of the candidate recognition paragraphs, determine a first overlap ratio between the candidate recognition paragraph and the text paragraph where each traffic-related entity in the traffic complaint text sample is located, and if the first overlap ratio is greater than a first preset overlap ratio, or if each of the first overlap ratios is less than a second preset overlap ratio, determine the candidate recognition paragraph as an initial recognition paragraph of the traffic complaint text sample, and the first preset overlap ratio is greater than the second preset overlap ratio;

[0016] For each of the traffic complaint text samples, adjusting each first recognized paragraph using the initial text description vector of the traffic complaint text sample to obtain a second recognized paragraph and a corresponding initial reasoning result includes:

[0017] Projecting the spatial distribution information of each first identified paragraph of the traffic complaint text sample onto the initial text description vector of the traffic complaint text sample to obtain an integrated text description vector;

[0018] Determine the second recognized paragraphs and the corresponding initial reasoning results corresponding to each first recognized paragraph of the traffic complaint text sample by using the integrated text description vector;

[0019] For each of the traffic complaint text samples, the initial reasoning result of the second recognition paragraph obtained through each round of optimization is used to determine the target reasoning result of each target recognition paragraph of the traffic complaint text sample, including:

[0020] For each initial recognition paragraph of the traffic complaint text sample, the integrated reasoning result corresponding to the initial recognition paragraph is determined by the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by each round of optimization except the last round of adjustment, and the target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph is determined by the integrated reasoning result and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the last round of adjustment.

[0021] In some embodiments, determining the total debugging error by using the target reasoning results and actual entity information of each target recognition segment includes:

[0022] Determine a first debugging error through the target reasoning results and actual entity information of each target recognition segment;

[0023] Determine the text paragraph where each traffic-related entity in each of the traffic complaint text samples is located, and determine the second debugging error through the text paragraph where each traffic-related entity in each of the traffic complaint text samples is located and each of the target recognition paragraphs;

[0024] A total debugging error is determined by the first debugging error and the second debugging error.

[0025] In some embodiments, determining the first debugging error by using the target reasoning result and the actual entity information of each target recognition segment includes:

[0026] Determine a first component error through the target reasoning result and the actual entity information of each target recognition segment, and determine a first debugging error through one or more of a second component error or a third component error and the first component error;

[0027] Determining the second component error includes:

[0028] Determine the actual entity information of each second recognition paragraph obtained in each round of optimization;

[0029] Determine the second composition error through the initial reasoning results and actual entity information of each second recognition segment obtained in each round of optimization;

[0030] Determining the third component error includes:

[0031] For each of the initially recognized paragraphs, determining an initial reasoning result of the initially recognized paragraph through the initially recognized paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located;

[0032] The actual entity information of each of the initially recognized paragraphs is determined, and the third composition error is determined through the initial reasoning results and the actual entity information of each of the initially recognized paragraphs.

[0033] In some embodiments, for each of the initially recognized paragraphs, determining the initial reasoning result of the initially recognized paragraph by using the initially recognized paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located includes:

[0034] If a first overlap rate between the initial recognition paragraph and a text paragraph in which one or more traffic-related entities in a corresponding traffic complaint text sample are located is greater than a first preset overlap rate, it is determined that the text network infers that the initial recognition paragraph includes a set type of traffic-related entity;

[0035] If the first overlap rate between the initial recognition paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located is less than the second preset overlap rate, it is determined that the text network infers that the initial recognition paragraph does not include a set type of traffic-related entity.

[0036] In some embodiments, determining the second debugging error through the text paragraph where each traffic-related entity is located in each of the traffic complaint text samples and each of the target identification paragraphs includes:

[0037] Determine a fourth component error through each target identification paragraph and the text paragraph where each traffic-related entity is located in each traffic complaint text sample, and determine a second debugging error through one or more of the fifth component error or the sixth component error and the fourth component error;

[0038] The fifth component error is determined by the second recognition paragraphs obtained from each round of optimization except the last round of adjustment and the text paragraph where each traffic-related entity is located in each traffic complaint text sample, and the sixth component error is determined by the initial recognition paragraphs and the text paragraph where each traffic-related entity is located in each traffic complaint text sample.

[0039] In some embodiments, for each target recognition segment, determining whether the target recognition segment includes a set type of traffic-related entity through a target reasoning result of the target recognition segment includes:

[0040] Determine a target entity description vector of a traffic-related entity corresponding to the target recognition segment;

[0041] Determine a commonality measurement result between the target entity description vector and a preset entity description vector of each set type of traffic-related entity in a description vector library, wherein each of the preset entity description vectors and the target entity description vector is determined by a pre-debugged traffic entity processing operator;

[0042] Whether the target recognition section includes a set type of traffic-related entity is determined by using the maximum commonality measurement result and the target reasoning result of the target recognition section.

[0043] In some embodiments, determining whether the target identification segment includes a set type of traffic-related entity by using the maximum commonality measurement result and the target reasoning result of the target identification segment includes:

[0044] If it is determined through the target reasoning result of the target recognition paragraph that the text keyword recognition network infers that the target recognition paragraph does not include a set type of traffic-related entity, and the maximum commonality measurement result is less than the first preset commonality measurement result, it is determined that the target recognition paragraph actually does not include a set type of traffic-related entity;

[0045] If it is determined through the target inference result of the target recognition paragraph that the text keyword recognition network infers that the target recognition paragraph includes a set type of traffic-related entities, and the maximum commonality measurement result is greater than the second preset commonality measurement result, then it is determined that the target recognition paragraph actually includes the set type of traffic-related entities, and the first preset commonality measurement result is not greater than the second preset commonality measurement result.

[0046] In a second aspect, an embodiment of the present application provides a traffic complaint handling device, including:

[0047] A text feature mining module is used to determine the initial text description vector of the traffic complaint text to be mined;

[0048] A text paragraph reasoning module is used to input the initial text description vector of the traffic complaint text to be mined into a text keyword recognition network to obtain one or more target recognition paragraphs and their target reasoning results; wherein each of the reasoning results is used to indicate whether the target recognition paragraph corresponding to the text keyword recognition network reasoning includes a set type of traffic-related entity;

[0049] A related entity recognition module, used for determining whether the target recognition section includes a set type of traffic related entity through the target reasoning result of each target recognition section;

[0050] The recognition network debugging module is used to debug the text keyword recognition network according to the following operations:

[0051] Inputting the initial text description vector of each traffic complaint text sample into the basic text network, obtaining target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample, each of which indicates whether the target recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity;

[0052] The total debugging error is determined by the target reasoning results and actual entity information of each target recognition paragraph, the basic text network is debugged by each traffic complaint text sample until the total debugging error meets the set requirements, and the debugging is stopped, and the text network when the debugging is stopped is determined as a text keyword recognition network, and each actual entity information is used to indicate whether the corresponding target recognition paragraph includes a set type of traffic-related entity; wherein the process of determining each target reasoning result by the basic text network includes:

[0053] Through the initial text description vector of each traffic complaint text sample, one or more initial recognition paragraphs of the traffic complaint text sample are repeatedly adjusted for a set number of rounds to obtain the corresponding target recognition paragraph, wherein each round of optimization includes: adjusting each first recognition paragraph through the initial text description vector of the traffic complaint text sample to obtain a second recognition paragraph and a corresponding initial reasoning result, each initial reasoning result indicates whether the second recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity, the first recognition paragraph corresponding to each round of optimization is the second recognition paragraph obtained by the previous round of adjustment, the first recognition paragraph corresponding to the first adjustment is each of the initial recognition paragraphs, and the second recognition paragraph obtained by the last round of adjustment is the target recognition paragraph of the traffic complaint text sample;

[0054] The target reasoning results of each target recognition paragraph of the traffic complaint text sample are determined by the initial reasoning results of the second recognition paragraph obtained through each round of optimization.

[0055] In a third aspect, the present application provides a traffic complaint handling device, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the steps in the above-described method when executing the computer program.

[0056] This application has at least the following beneficial effects:

[0057] Through the traffic complaint processing method based on text mining provided by the embodiment of the present application, each initial recognition paragraph of the traffic complaint text sample can be repeatedly adjusted through the initial text description vector of the traffic complaint text sample, so that the target recognition paragraph finally obtained can accurately identify the traffic-related entities in the traffic complaint text sample. At the same time, in each repeated adjustment, the initial reasoning result of the recognition paragraph obtained by each repeated adjustment can also be determined through the initial text description vector of the traffic complaint text sample, so as to determine the target reasoning result of the final target recognition paragraph through the initial reasoning result of the recognition paragraph obtained by each repeated adjustment, so as to increase the accuracy of the target reasoning result of each target recognition paragraph. Through this debugging method, the text keyword recognition network of the set type of traffic-related entities in the text can be accurately obtained. In addition, for the traffic complaint text to be mined, after the text keyword recognition network obtained through debugging determines the target recognition paragraph of the traffic complaint text to be mined, and the target reasoning result indicating whether each target recognition paragraph includes a set type of traffic-related entity is inferred by the text keyword recognition network, it can be determined whether the traffic-related entity corresponding to each target recognition paragraph is a set type of traffic-related entity through the entity description vector of the traffic-related entity corresponding to each target recognition paragraph and the corresponding target reasoning result, thereby ensuring the recognition accuracy of the set type of traffic-related entities.

[0058] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0060] Figure 1 A schematic diagram of the debugging process of a text keyword recognition network provided in an embodiment of the present application.

[0061] Figure 2A schematic diagram of the implementation flow of a traffic complaint handling method based on text mining provided in an embodiment of the present application.

[0062] Figure 3 A schematic diagram of the structure of a traffic complaint handling device provided in an embodiment of the present application.

[0063] Figure 4 A hardware entity diagram of a traffic complaint handling device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0065] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "first / second / third" involved are merely used to distinguish similar objects and do not represent a specific order for the objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing this application and are not intended to limit this application.

[0067] The embodiment of the present application provides a traffic complaint processing method based on text mining, which can be executed by a processor of a traffic complaint processing device. The traffic complaint processing device can refer to a server, a laptop, a tablet computer, a desktop computer, a smart TV, a mobile device (such as a mobile phone, a portable video player, a personal digital assistant, a dedicated messaging device, a portable gaming device) and other devices with data processing capabilities.

[0068] The text mining-based traffic complaint handling method provided in the embodiment of the present application uses a text keyword recognition network to identify key complaint information. In order to facilitate a better understanding of the solution of the present application, the debugging process of the text keyword recognition network is first introduced below, and then the text mining-based traffic complaint handling method based on the text keyword recognition network is introduced.

[0069] Please refer to Figure 1 Specifically, the debugging process of the text keyword recognition network includes the following steps:

[0070] Step S110, inputting the initial text description vector of each traffic complaint text sample into the basic text network to obtain the target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample.

[0071] The traffic complaint text sample is the training data for debugging. As an implementation method, after determining multiple traffic complaint text samples, the text description vector of each traffic complaint text sample can be extracted by the pre-debugging text description vector extraction operator to obtain the initial text description vector of each traffic complaint text sample. The text description vector extraction operator is a feature extraction operator, which can be a convolutional neural network (such as ResNet), which is obtained by pre-debugging (pre-train). It can be understood that before extracting features from the text, a general pre-processing method can be used for pre-processing, such as removing special characters in the text, segmenting the text, removing stop words, and performing feature selection based on the bag of words model (Bag of Words), TF-IDF, Word2Vec, etc. Finally, based on the feature extraction operator, the text is converted into a vector form using methods such as One-Hot Encoding and Word Embedding to obtain an initial text description vector. In other words, the initial text description vector is a vector that describes the features of the traffic complaint text sample.

[0072] As a specific architecture example of a feature extraction operator, ResNet includes cascaded pooling layers, N residual layers (the setting of each residual layer is configured according to actual needs and is not limited), pooling layers, and dense layers (also known as affine layers, fully connected layers). For each traffic complaint text sample, the traffic complaint text sample can be pooled through the first pooling layer to complete downsampling, and the pooled features are subjected to residual processing through multiple cascaded residual layers to obtain hidden layer features, and then the hidden layer features are processed through the second pooling layer and the dense layer to obtain the initial text description vector of the traffic complaint text sample.

[0073] As an implementation method, for each traffic complaint text sample, the initial text description vector of the traffic complaint text sample is input into the basic text network (i.e., the prepared initial neural network to be debugged), and one or more target recognition paragraphs and corresponding target reasoning results of the traffic complaint text sample are obtained. Among them, the target reasoning result of each target recognition paragraph indicates whether the target recognition paragraph includes a set type of traffic-related entity by the text network reasoning. Among them, for each traffic complaint text sample, any two target recognition paragraphs of the traffic complaint text sample may correspond to the same or different traffic-related entities. In other words, each traffic-related entity in the traffic complaint text sample may correspond to one or more target recognition paragraphs. Traffic-related entities are entity information associated with traffic information in the traffic complaint text, and their types are related to the set types. The set types are the types of traffic information. For example, the set types may at least include community names, road names, intersection names, etc. Taking the community name as an example, the corresponding traffic-related entity is the entity information related to the community, such as "High-tech Zone Experimental Junior High School". The target recognition paragraph is the text position information in the traffic complaint text, or the target recognition area, the target recognition position, for example, it may indicate a certain paragraph, a certain sentence, a certain character interval in a certain sentence, etc.

[0074] For any target recognition paragraph of each traffic complaint text sample, the traffic-related entity corresponding to the target recognition paragraph may be completely located in the target recognition paragraph, or partially located in the target recognition paragraph.

[0075] The target reasoning results of different target recognition paragraphs corresponding to the same traffic-related entity may be the same or different. For example, in the initial stage of network debugging, the target reasoning results of different target recognition paragraphs corresponding to the same traffic-related entity are likely to be erroneous. In the later stage of network debugging, the target reasoning results of different target recognition paragraphs corresponding to the same traffic-related entity are likely to be consistent. The method for determining the target reasoning results of each target paragraph by the basic text network may specifically include:

[0076] Step S210 , repeatedly adjusting one or more initial recognition paragraphs of the traffic complaint text sample for a set number of rounds using the initial text description vector of each traffic complaint text sample to obtain a corresponding target recognition paragraph.

[0077] As an implementation method, for each traffic complaint text sample, after the initial text description vector of the traffic complaint text sample is input into the basic text network, the basic text network can process the initial text description vector of each traffic complaint text sample through a paragraph determination operator to obtain one or more initial recognition paragraphs of the traffic complaint text sample. The paragraph determination operator can be implemented using an attention mechanism (Attention-based Mechanisms) network to dynamically assign importance weights to different positions based on the input context information. By calculating the correlation score between each position and the remaining positions, and then normalizing these scores into a probability distribution for weighted aggregation of the representation of the input sequence, the text paragraph with the highest probability is determined as the target recognition paragraph.

[0078] For each traffic complaint text sample, since each vector element position of the initial text description vector of the traffic complaint text sample corresponds to a text paragraph in the traffic complaint text sample, after the initial text description vector of the traffic complaint text sample is input into the basic text network, multiple text paragraphs of the traffic complaint text sample can be determined by the initial text description vector of the traffic complaint text sample. Further, for each text paragraph, a random position of the text paragraph (such as the starting character, the ending character) is used as a reference character, and by extending the selection range (such as selecting 3 characters before and after the reference character, or selecting 3 characters before the reference character, or selecting 3 characters after the reference character, without specific limitation), multiple target text paragraphs are determined in the corresponding traffic complaint text sample as candidate recognition paragraphs, which can be regarded as a text box.

[0079] After all candidate recognition paragraphs corresponding to all traffic complaint text samples are determined, each candidate recognition paragraph is used as an alternative debugging sample to be used in this round of network debugging.

[0080] For each traffic complaint text sample, after determining all the candidate identification paragraphs corresponding to the traffic complaint text sample, determine the text paragraph of each traffic-related entity in the traffic complaint text sample. Among them, the text paragraph of each traffic-related entity in the traffic complaint text sample is a paragraph of the set text length that completely covers the traffic-related entity. For each candidate identification paragraph of each traffic complaint text sample, determine the overlap rate between the candidate identification paragraph and the text paragraph of each traffic-related entity in the traffic complaint text sample (as the first overlap rate, that is, the intersection-and-union ratio between the text paragraphs, also known as the IOU value, and the overlap data compared is the segmentation in the paragraph). If the first overlap rates corresponding to the candidate identification paragraphs have a first overlap rate greater than the first preset overlap rate, or the first overlap rates corresponding to the candidate identification paragraphs are all less than the second preset overlap rate, the candidate identification paragraph is determined as an initial identification paragraph of the traffic complaint text sample. Among them, the first preset overlap rate is greater than the second preset overlap rate, and the specific values ​​of the first preset overlap rate and the second preset overlap rate are determined according to actual needs. For example, the first preset overlap rate is set to 60%, and the second preset overlap rate is set to 40%.

[0081] For each traffic complaint text sample, the step of determining the initial recognition paragraph from each candidate recognition paragraph of the traffic complaint text sample can be regarded as the process of determining the debugging sample used in this round of network debugging from each candidate debugging sample. For each candidate recognition paragraph of each traffic complaint text sample, if the first overlap rates corresponding to the candidate recognition paragraphs have a first overlap rate greater than the first preset overlap rate, the candidate recognition paragraph is regarded as a positive sample (i.e., a positive sample); if the first overlap rates corresponding to the candidate recognition paragraphs are all less than the second preset overlap rate, the candidate recognition paragraph is regarded as a negative sample (i.e., a negative sample).

[0082] As an implementation method, for each traffic complaint text sample, in the process of repeatedly adjusting the traffic complaint text sample and the initial recognition paragraph through the initial text description vector of the traffic complaint text sample for a set number of rounds to obtain the corresponding target recognition paragraph, for each round of optimization, the initial text description vector of the traffic complaint text sample is used to adjust each first recognition paragraph to obtain the corresponding second recognition paragraph and the initial inference result corresponding to the second recognition paragraph.

[0083] Each initial inference result indicates whether the second recognition paragraph corresponding to the text network inference includes a set type of traffic-related entity. The first recognition paragraph corresponding to each round of optimization is each second recognition paragraph obtained by the previous round of adjustment. The first recognition paragraph corresponding to the first adjustment is each initial recognition paragraph of the traffic complaint text sample. The second recognition paragraph obtained by the last round of adjustment is determined as the target recognition paragraph of the traffic complaint text sample. The initial inference result of each second recognition paragraph obtained by each round of optimization is expressed as the confidence that the second recognition paragraph includes a set type of traffic-related entity. When the confidence corresponding to the initial inference result is greater than the first preset confidence, it indicates that the second recognition paragraph corresponding to the text network inference includes a set type of traffic-related entity. Otherwise, it indicates that the second recognition paragraph corresponding to the text network inference does not include a set type of traffic-related entity.

[0084] For each round of optimization, the spatial distribution information of each first recognized paragraph of the traffic complaint text sample is projected (i.e., mapped) to the initial text description vector of the traffic complaint text sample to obtain the integrated text description vector of the traffic complaint text sample, and then the second recognized paragraph corresponding to each first recognized paragraph of the traffic complaint text sample and the corresponding initial reasoning result are determined through the integrated text description vector. Among them, the spatial distribution information is the position information of the recognized paragraph, and the position information is mapped to the vector so that the initial text description vector incorporates the position feature information to obtain the integrated text description vector.

[0085] The basic text network includes a plurality of adjustment operators connected in sequence, and each round of optimization is performed by the adjustment operators of the corresponding layer. In one example, after obtaining each second recognition paragraph through the adjustment operator of the previous layer, any adjustment operator of the basic text network loads the initial text description vector of the traffic complaint text sample and each first recognition paragraph obtained by the adjustment operator of the previous layer into the description vector fusion operator in the adjustment operator (used to complete feature fusion) to obtain the integrated text description vector of the traffic complaint text sample, and then processes the integrated text description vector through the description vector processing operator to obtain the second recognition paragraph corresponding to each first recognition paragraph and the corresponding initial reasoning result. The second recognition paragraph currently obtained is the first recognition paragraph corresponding to the adjustment operator of the next layer. The parameters of the adjustment operators at each level can be consistent or inconsistent.

[0086] Step S220 , determining target reasoning results of each target recognition paragraph of the traffic complaint text sample through the initial reasoning results of the second recognition paragraph obtained in each round of optimization.

[0087] As an implementation method, for each traffic complaint text sample, each target recognition paragraph of the traffic complaint text sample is obtained, and for each initial recognition paragraph of the traffic complaint text sample, the initial reasoning results of the second recognition paragraph corresponding to the initial recognition paragraph obtained in each round of optimization are integrated to obtain the target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph. For each traffic complaint text sample, after obtaining each target recognition paragraph of the traffic complaint text sample, the sum of the weights of the initial reasoning results of the second recognition paragraph corresponding to the initial recognition paragraph obtained in each round of optimization is determined, and the sum of the weights is determined as the target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph. Similarly, the target reasoning result of each target recognition paragraph can be expressed as the confidence that the target recognition paragraph includes a set type of traffic-related entity. The weights corresponding to the second recognition paragraphs obtained by any two adjustments are the same or different.

[0088] As an implementation mode, when determining the target reasoning result of each target recognition paragraph of the traffic complaint text sample, for each initial recognition paragraph of the traffic complaint text sample, the initial reasoning result corresponding to the initial recognition paragraph is determined by the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by each round of optimization except the last round of adjustment. The sum of the weights of the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by each round of optimization except the last round of adjustment can be determined as the integrated reasoning result corresponding to the initial recognition paragraph. The weights corresponding to the second recognition paragraph obtained by any two adjustments can be the same or different. Afterwards, the integrated reasoning result corresponding to the initial recognition paragraph and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the last round of adjustment can be integrated to obtain the target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph. The target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph can be determined by the sum of the weights of the integrated reasoning result corresponding to the initial recognition paragraph and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the last round of adjustment. Among them, the weights of the integrated reasoning result and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the last round of adjustment can be the same or different. Among them, for the initial reasoning results of the second recognized paragraph corresponding to the initial recognized paragraph obtained by each round of optimization except the last round of adjustment, in order to improve the recognition quality of the initial reasoning results of the basic text network, the initial reasoning results can be deeply processed by the deep processing operator to obtain the first reasoning result. The integrated reasoning result of the initial recognized paragraph is determined by determining the first reasoning result of the second recognized paragraph corresponding to the initial recognized paragraph obtained by each round of optimization except the last round of adjustment. Among them, each round of optimization can use the same or different deep processing operators, and the operator parameters of each deep processing operator are not limited.

[0089] As a scenario for determining the integrated reasoning results, for each traffic complaint text sample, if the initial recognition paragraph of the traffic complaint text sample is adjusted for x rounds to obtain the target recognition paragraph, then for each initial recognition paragraph, the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the first adjustment can be loaded into the deep processing operator to obtain the first reasoning result corresponding to this round of adjustment, and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the second round of adjustment can be loaded into the deep processing operator to obtain the first reasoning result corresponding to the second adjustment, and so on, the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the x-1th round of adjustment can be loaded into the deep processing operator to obtain the first reasoning result corresponding to the second adjustment. Then, the first reasoning results of the second recognition paragraph corresponding to the initial recognition paragraph obtained from the first adjustment to the x-1th round of adjustment can be integrated to obtain the integrated reasoning result of the initial recognition paragraph.

[0090] In the process of processing the initial reasoning result by the deep processing operator, the reasoning description vector used to characterize the initial reasoning result can be processed by the deep processing operator, and the reasoning description vector of the initial reasoning result is the input of the dense layer for determining the initial reasoning result.

[0091] As an architectural example of a deep processing operator, each deep processing operator may include n convolutional layers and one dense layer, wherein every two convolutional layers are identically connected across layers, the operator parameters of each deep processing operator are independent of each other, and after each initial inference result is loaded into the deep processing operator, a corresponding first inference result is output based on the deep processing operator.

[0092] The following is a more in-depth introduction to the network structure of the basic text network. For each traffic complaint text sample, after determining the initial text description vector of the traffic complaint text sample, the initial text description vector of the traffic complaint text sample is input into the basic text network.

[0093] Among them, the basic text network processes the initial text description vector through the attention network to obtain multiple initial recognition paragraphs, and adjusts the initial recognition paragraphs through multiple layers of adjustment operators to obtain the target recognition paragraphs corresponding to each initial recognition paragraph. Among them, the basic text network projects the spatial distribution information of the initial recognition paragraphs to the initial text description vector to obtain the integrated text description vector, and processes the integrated text according to the first layer of adjustment operators to obtain the second recognition paragraphs corresponding to each initial recognition paragraph and the corresponding initial reasoning results. The spatial distribution information of each second recognition paragraph obtained by adjusting the first layer of adjustment operators is projected to the initial text description vector to obtain the integrated text description vector, and the integrated text description vector obtained in this round is processed by the second layer of adjustment operators to obtain a new second recognition paragraph and the corresponding initial reasoning results. The spatial distribution information of each second recognition paragraph obtained by the second layer of adjustment operators is projected to the initial text description vector to obtain the integrated text description vector, and the integrated text description vector obtained in this round is processed by the third layer of adjustment operators to obtain a new second recognition paragraph and the corresponding initial reasoning results. The spatial distribution information of each second recognition paragraph obtained by the third-level adjustment is projected onto the initial text description vector to obtain the integrated text description vector. The integrated text description vector obtained in this round is processed by the last-level adjustment operator to obtain the target recognition paragraph and the corresponding initial reasoning result corresponding to each initial recognition paragraph. For each initial recognition paragraph, the basic text network uses the initial reasoning results of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the first-level adjustment operator, the second-level adjustment operator and the third-level adjustment operator to process the corresponding initial reasoning results respectively with the deep processing operator corresponding to each layer of adjustment operator to obtain the corresponding first reasoning result, and obtains the integrated reasoning result corresponding to the initial recognition paragraph through each first reasoning result. Afterwards, the basic text network integrates the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the fourth-level adjustment operator with the integrated reasoning result corresponding to the initial recognition paragraph to obtain the target reasoning result of the target reasoning paragraph corresponding to the initial reasoning result. Afterwards, the target reasoning result of each target recognition paragraph is used to determine whether the target recognition paragraph inferred by the basic text network includes a set type of traffic-related entity.

[0094] Step S120, determine the total debugging error through the target reasoning results and actual entity information of each target recognition paragraph, debug the basic text network through each traffic complaint text sample until the total debugging error meets the set requirements, and stop debugging, and determine the text network when debugging is stopped as the text keyword recognition network.

[0095] Each traffic-related entity of each traffic complaint text sample has entity annotation information, which is used to characterize whether the corresponding traffic-related entity is a traffic-related entity of a set type. Then, after determining each target identification paragraph, the actual entity information of each target identification paragraph is determined by the entity annotation information of each traffic-related entity in each traffic complaint text sample. In an embodiment of the present application, each target reasoning result is the confidence that the corresponding target identification paragraph includes a traffic-related entity of a set type. If the confidence corresponding to the target reasoning result is greater than the first preset confidence, it means that the network reasoning corresponds to the target identification paragraph including the traffic-related entity of the set type. On the contrary, it means that the network reasoning corresponds to the second identification paragraph does not include the traffic-related entity of the set type. As an implementation method, when determining the total debugging error, the first debugging error is determined by the target reasoning result and the actual entity information of each target identification paragraph, and the text paragraph where each traffic-related entity in each traffic complaint text sample is located is determined, and then the second debugging error is determined by the text paragraph where each traffic-related entity in each traffic complaint text sample is located and each target identification paragraph. Among them, the actual entity information of each target identification paragraph indicates whether the corresponding target identification paragraph includes a traffic-related entity of a set type.

[0096] Afterwards, the total debugging error is calculated by the first debugging error and the second debugging error, for example, the result after adding the first debugging error and the second debugging error is used as the total debugging error, or the weighted sum of the first debugging error and the second debugging error is determined as the total debugging error. When determining the first debugging error, for example, the cross entropy error between the target reasoning result and the actual entity information of each target recognition paragraph is calculated as the first debugging error. When determining the first debugging error, the cross entropy error between the target reasoning result and the actual entity information of each target recognition paragraph can also be calculated as the first component error. In addition, the actual entity information of each second recognition paragraph obtained in each round of optimization can also be determined, for example, the actual entity information of each second recognition paragraph obtained in each round of optimization is determined by the entity annotation information of each traffic-related entity in each traffic complaint text sample. Then, the second component error is determined by the initial reasoning result and the actual entity information of each second recognition paragraph obtained in each round of optimization. Among them, for example, the cross entropy error between the initial reasoning result and the actual entity information of each second recognition paragraph obtained in each round of optimization is determined as the second component error. The actual entity information corresponding to each second recognition paragraph indicates whether the second recognition paragraph includes a set type of traffic-related entity.

[0097] Furthermore, the actual entity information of each initially recognized paragraph can be determined, for example, by using the entity annotation information of each traffic-related entity in each traffic complaint text sample to determine the actual entity information of each initially recognized paragraph. Then, the third component error is determined by the initial reasoning results and actual entity information of each initially recognized paragraph. Among them, for example, the cross entropy error of the initial reasoning results and actual entity information of each initially recognized paragraph is calculated as the third component error. Among them, the actual entity information corresponding to each initially recognized paragraph indicates whether the initially recognized paragraph includes a set type of traffic-related entity, and the initial reasoning result of each initially recognized paragraph indicates whether the initially recognized paragraph includes a set type of traffic-related entity by text network reasoning.

[0098] For each initially recognized paragraph, it can be understood that the first overlap rate between each initially recognized paragraph and the area where each traffic-related entity in the corresponding traffic complaint text sample is located has a first overlap rate greater than the first preset overlap rate, or each first overlap rate is less than the second preset overlap rate. Then for each initially recognized paragraph, if the first overlap rate between the initially recognized paragraph and the text paragraph where one or more traffic-related entities in the corresponding traffic complaint text sample are located is greater than the first preset overlap rate, it is determined that the text network inference initially recognized paragraph includes a set type of traffic-related entity; if the first overlap rate between the initially recognized paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located is less than the second preset overlap rate, it is determined that the text network inference initially recognized paragraph does not include a set type of traffic-related entity. Among them, the first preset overlap rate is greater than the second preset overlap rate.

[0099] When determining the first debugging error, the first debugging error is determined by one or more of the second component error or the third component error and the first component error, and the component error can be understood as a sub-error. For example, the sum of the component errors is used as the first debugging error, such as adding the first component error, the second component error and the third component error, and taking the sum as the first debugging error. The weighted sum of the component errors can also be used as the first debugging error, such as weighted summing the first component error, the second component error and the third component error, and taking the result as the first debugging error. When determining the second debugging error, for example, the position-sensitive weighted loss (Position-Sensitive Weighted Loss) of the text paragraph where each traffic-related entity is located in each target recognition paragraph and each traffic complaint text sample is calculated, and the loss is determined as the second debugging error.

[0100] When determining the second debugging error, the position-sensitive weighted loss of each target recognition paragraph and the text paragraph where each traffic-related entity in each traffic complaint text sample is located can also be determined and used as the fourth component error. In addition, the fifth component error can be determined by each second recognition paragraph obtained by each round of optimization except the last round of adjustment and the text paragraph where each traffic-related entity in each traffic complaint text sample is located. For example, the position-sensitive weighted loss of each second recognition paragraph obtained by each round of optimization except the last round of adjustment and the text paragraph where each traffic-related entity in each traffic complaint text sample is located is used as the fifth component error. In addition, the sixth component error can be determined by each initial recognition paragraph and the text paragraph where each traffic-related entity in each traffic complaint text sample is located, for example, the position-sensitive weighted loss of each initial recognition paragraph and the text paragraph where each traffic-related entity in each traffic complaint text sample is located is used as the sixth component error.

[0101] When determining the second debugging error, for example, the second debugging error is determined by one or more of the fifth component error or the sixth component error and the fourth component error. The sum of the component errors can be used as the second debugging error, for example, the sum of the fourth component error, the fifth component error and the sixth component error can be used as the first debugging error. Alternatively, the weighted sum of the component errors can be used as the second debugging error, for example, the fourth component error, the fifth component error and the sixth component error can be weighted summed, and the weighted sum can be used as the second debugging error.

[0102] For any second recognized paragraph in any round of adjustment, the overlap rate (deemed as the second overlap rate) between the second recognized paragraph and the paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located is determined. If among the second overlap rates corresponding to the second recognized paragraph, there is a second overlap rate greater than the third preset overlap rate, then the second recognized paragraph is regarded as a positive example; if the second overlap rates corresponding to the second recognized paragraph are all less than the fourth preset overlap rate, the second recognized paragraph is regarded as a negative example.

[0103] As an implementation method, after determining the total debugging error corresponding to each debugging, the basic text network is repeatedly debugged through the initial text description vector of each traffic complaint text sample until the total debugging error meets the set requirements (for example, the total debugging error converges), and then the debugging is stopped, and the text network obtained by stopping the debugging is determined as the text keyword recognition network.

[0104] The above content completes the introduction of the debugging process of the text keyword recognition network. After the text keyword recognition network is obtained based on the above embodiment, the embodiment of the present application implements a traffic complaint processing method based on text mining based on the text keyword recognition network. The following introduces the operation process of the traffic complaint processing method based on text mining. Please refer to Figure 2The traffic complaint handling method based on text mining provided in the embodiment of the present application includes the following steps:

[0105] Step S310, determining the initial text description vector of the traffic complaint text to be mined, inputting the initial text description vector of the traffic complaint text to be mined into a text keyword recognition network, and obtaining one or more target recognition paragraphs and their target reasoning results.

[0106] As an implementation method, the text keyword recognition network determines one or more target recognition paragraphs of the traffic complaint text to be mined and their target reasoning results through the initial text description vector of the traffic complaint text to be mined, and refers to the above step S110. Among them, the text keyword recognition network is a text network obtained by debugging the text keyword recognition network debugging method provided in the embodiment of the present application. Each target reasoning result indicates whether the target recognition paragraph corresponding to the text keyword recognition network reasoning includes a set type of traffic-related entity.

[0107] Step S320 , determining whether the target recognition section includes a set type of traffic-related entity through the target inference result of each target recognition section.

[0108] As an implementation method, each target reasoning result is, for example, a confidence level that the corresponding target recognition segment includes a set type of traffic-related entity. In other embodiments, it may also be a probability. Then, for each target recognition segment, if the confidence level corresponding to the target reasoning result is greater than the first preset confidence level, it means that the text keyword recognition network reasoning corresponding to the target recognition segment includes a set type of traffic-related entity. Conversely, it means that the text keyword recognition network reasoning corresponding to the second recognition segment does not include a set type of traffic-related entity.

[0109] As an implementation method, for each target identification paragraph, when determining whether the target identification paragraph includes a set type of traffic-related entity, the target entity description vector (i.e., a text feature vector used to describe the traffic-related entity) of the traffic-related entity corresponding to the target identification paragraph can be determined, and then the commonality measurement result (used to describe the degree of similarity between the two vectors) between the target entity description vector and the preset entity description vector of each set type of traffic-related entity in the description vector library (preset feature library) is determined. Specifically, the commonality measurement result can be expressed by calculating the vector distance of the two vectors, such as based on the Euclidean distance as the commonality measurement result, the smaller the distance, the larger the commonality measurement result. When determining the traffic-related entity corresponding to the target identification paragraph, the target identification paragraph is framed from the traffic complaint text to be mined to form a text frame to obtain the traffic-related entity corresponding to the target identification paragraph. Among them, each preset entity description vector and the target entity description vector corresponding to the target identification paragraph are determined by the pre-debugged traffic entity processing operator.

[0110] For each set type of traffic-related entity and any traffic-related entity corresponding to the target identification section, the traffic-related entity is processed by a pre-debugged traffic entity processing operator to obtain an entity description vector. If the traffic-related entity is a traffic-related entity of the set type, the obtained entity description vector is determined to be a preset entity description vector of the traffic-related entity of the set type; if the traffic-related entity is a traffic-related entity corresponding to the target identification section, the obtained entity description vector is determined to be a target entity description vector of the traffic-related entity corresponding to the target identification section.

[0111] Then, for each preset target recognition paragraph, determine the maximum commonality measurement result among the commonality measurement results of the corresponding target entity description vector and the preset entity description vector of each set type of traffic-related entity in the description vector library, and determine whether the target recognition paragraph includes the set type of traffic-related entity through the maximum commonality measurement result and the target reasoning result of the target recognition paragraph. If it is determined through the target reasoning result of the target recognition paragraph that the text keyword recognition network reasoning target recognition paragraph does not include the set type of traffic-related entity, and the maximum commonality measurement result is less than the first preset commonality measurement result, then it is determined that the target recognition paragraph does not include the set type of traffic-related entity; if it is determined through the target reasoning result of the target recognition paragraph that the text keyword recognition network reasoning target recognition paragraph includes the set type of traffic-related entity, and the maximum commonality measurement result is greater than the second preset commonality measurement result, then it is determined that the target recognition paragraph includes the set type of traffic-related entity. The first preset commonality measurement result is not greater than the second preset commonality measurement result.

[0112] For other scenarios, the traffic-related entity corresponding to the target recognition segment may be uploaded to the expert system, and whether the traffic-related entity is a set type of traffic-related entity may be manually determined based on the expert system.

[0113] As an implementation method, if each target reasoning result indicates the confidence that the corresponding target recognition segment includes a set type of traffic-related entity, for each target recognition segment, if the confidence corresponding to the target recognition segment is not less than the second preset confidence, it is determined that the target recognition segment includes the set type of traffic-related entity. If the confidence corresponding to the target recognition segment is less than the second preset confidence, it means that the target recognition segment may not include the set type of traffic-related entity, then the commonality measurement result of the target entity description vector of the traffic-related entity corresponding to the target recognition segment and the preset entity description vector of each set type of traffic-related entity in the description vector library is determined, and whether the target recognition segment includes the set type of traffic-related entity is determined through the maximum commonality measurement result and the target reasoning result of the target recognition segment. The second preset confidence is greater than the first preset confidence.

[0114] For other scenarios, the traffic-related entity corresponding to the target recognition segment may be uploaded to the expert system, and whether the traffic-related entity is a set type of traffic-related entity may be manually determined based on the expert system.

[0115] In an embodiment of the present application, the pre-debugged traffic entity processing operator is debugged through the following operations, specifically, including: obtaining a debugging learning sample library, the debugging learning sample library including a plurality of traffic-related entity samples, each traffic-related entity sample annotated with entity annotation information, each entity annotation information representing whether the corresponding traffic-related entity sample is a set type of traffic-related entity; each traffic complaint text sample is a text with a preset text length. Each traffic-related entity sample is loaded into the initial traffic entity processing operator to obtain the sample entity description vector of each traffic-related entity sample, and each sample entity description vector is loaded into the first dense layer to obtain the classification mapping information of each traffic-related entity sample, each classification mapping information represents whether the corresponding traffic-related entity sample is a set type of traffic-related entity; the total debugging error is calculated through the real type of each traffic-related entity sample (whether it belongs to the set type of traffic-related entity) and the corresponding classification mapping information, and the initial traffic entity processing operator and the first dense layer are debugged through each traffic-related entity sample until the total debugging error meets the stopping requirement, and the debugging is stopped, and the initial traffic entity processing operator obtained by stopping the debugging is determined as the pre-debugging traffic entity processing operator.

[0116] The above initial traffic entity processing operator is obtained by combining the description vector processing operator of the binary traffic-related entity classification model in the related art with the second dense layer.

[0117] As a structural example of the initial traffic entity processing operator, a currently known two-class traffic-related entity classification model is formed by describing a vector processing operator and a first dense layer, and a second dense layer is added between the description vector processing operator and the first dense layer, and the description vector processing operator and the second dense layer are determined as the initial traffic entity processing operator. The initial traffic entity processing operator and the first dense layer are debugged through each traffic-related entity sample, and the initial traffic entity processing operator when the debugging stops is determined as the final pre-debugging traffic entity processing operator.

[0118] In summary, through the debugging process of the text keyword recognition network in the embodiment of the present application, each initial recognition paragraph of the traffic complaint text sample can be repeatedly adjusted through the initial text description vector of the traffic complaint text sample, so that the target recognition paragraph finally obtained can accurately recognize the traffic-related entities in the traffic complaint text sample. At the same time, in each repeated adjustment, the initial reasoning result of the recognition paragraph obtained by each repeated adjustment is determined by the initial text description vector of the traffic complaint text sample, and the target reasoning result of the final target recognition paragraph is determined by the initial reasoning result of the recognition paragraph obtained by each repeated adjustment, so as to increase the accuracy of the target reasoning result of each target recognition paragraph. Through this debugging method, a text keyword recognition network of a set type of traffic-related entity in the text can be accurately obtained. In addition, for the traffic complaint text to be mined, after the text keyword recognition network obtained through debugging determines the target recognition paragraph of the traffic complaint text to be mined, and indicates the target reasoning result of the text keyword recognition network to infer whether each target recognition paragraph includes a set type of traffic-related entity, because the text keyword recognition network has the probability of reasoning deviation, then the entity description vector of the traffic-related entity corresponding to each target recognition paragraph and the corresponding target reasoning result can be used to determine whether the traffic-related entity corresponding to each target recognition paragraph is a set type of traffic-related entity, thereby ensuring the recognition accuracy of the set type of traffic-related entities.

[0119] Based on the foregoing embodiments, the embodiments of the present application provide a traffic complaint handling device, and the various units included in the device, as well as the various modules included in each unit, can be implemented by a processor in a computer device; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0120] Figure 3 A schematic diagram of the structure of a traffic complaint handling device 300 provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the traffic complaint handling device 300 includes:

[0121] A text feature mining module 310 is used to determine an initial text description vector of a traffic complaint text to be mined;

[0122] The text paragraph reasoning module 320 is used to input the initial text description vector of the traffic complaint text to be mined into the text keyword recognition network to obtain one or more target recognition paragraphs and their target reasoning results; wherein each of the reasoning results is used to indicate whether the target recognition paragraph corresponding to the text keyword recognition network reasoning includes a set type of traffic-related entity;

[0123] A related entity recognition module 330, configured to determine whether the target recognition section includes a set type of traffic related entity through the target reasoning result of each target recognition section;

[0124] The recognition network debugging module 340 is used to debug the text keyword recognition network according to the following operations:

[0125] Inputting the initial text description vector of each traffic complaint text sample into the basic text network, obtaining target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample, each of which indicates whether the target recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity;

[0126] The total debugging error is determined by the target reasoning results and actual entity information of each target recognition paragraph, the basic text network is debugged by each traffic complaint text sample until the total debugging error meets the set requirements, and the debugging is stopped, and the text network when the debugging is stopped is determined as a text keyword recognition network, and each actual entity information is used to indicate whether the corresponding target recognition paragraph includes a set type of traffic-related entity; wherein the process of determining each target reasoning result by the basic text network includes:

[0127] Through the initial text description vector of each traffic complaint text sample, one or more initial recognition paragraphs of the traffic complaint text sample are repeatedly adjusted for a set number of rounds to obtain the corresponding target recognition paragraph, wherein each round of optimization includes: adjusting each first recognition paragraph through the initial text description vector of the traffic complaint text sample to obtain a second recognition paragraph and a corresponding initial reasoning result, each initial reasoning result indicates whether the second recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity, the first recognition paragraph corresponding to each round of optimization is the second recognition paragraph obtained by the previous round of adjustment, the first recognition paragraph corresponding to the first adjustment is each of the initial recognition paragraphs, and the second recognition paragraph obtained by the last round of adjustment is the target recognition paragraph of the traffic complaint text sample;

[0128] The target reasoning results of each target recognition paragraph of the traffic complaint text sample are determined by the initial reasoning results of the second recognition paragraph obtained through each round of optimization.

[0129] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present application can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0130] It should be noted that in the embodiment of the present application, if the above-mentioned traffic complaint handling method based on text mining is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific hardware, software or firmware, or any combination of hardware, software, and firmware.

[0131] An embodiment of the present application provides a traffic complaint handling device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0132] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, some or all of the steps in the above method are implemented. The computer-readable storage medium can be transient or non-transient.

[0133] An embodiment of the present application provides a computer program, including a computer-readable code. When the computer-readable code is run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0134] The embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium, and in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0135] It should be noted here that the description of the various embodiments above tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. The description of the above device, storage medium, computer program and computer program product embodiments is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiments of the device, storage medium, computer program and computer program product of this application, please refer to the description of the method embodiment of this application for understanding.

[0136] Figure 4 A hardware entity diagram of a traffic complaint handling device provided in an embodiment of the present application, such as Figure 4 As shown, the hardware entity of the traffic complaint handling device 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program.

[0137] The memory 1002 stores a computer program that can be run on the processor. The memory 1002 is configured to store instructions and applications executable by the processor 1001. It can also cache data to be processed or processed by the processor 1001 and each module in the traffic complaint handling device 1000 (for example, image data, audio data, voice communication data, and video communication data). It can be implemented through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0138] When the processor 1001 executes the program, the steps of any of the above-mentioned traffic complaint handling methods based on text mining are implemented. The processor 1001 generally controls the overall operation of the traffic complaint handling device 1000.

[0139] An embodiment of the present application provides a computer storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps of the traffic complaint handling method based on text mining as described in any of the above embodiments.

[0140] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For the technical details not disclosed in the storage medium and device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding. The above processor can be at least one of a target application integrated circuit (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, and a microprocessor. It can be understood that the electronic device that realizes the above processor function can also be other, and the embodiments of the present application are not specifically limited.

[0141] The above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM) and the like; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0142] It should be understood that the "one embodiment" or "one embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in one embodiment" appearing in various places throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of each step / process mentioned above does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The above-mentioned sequence number of the embodiment of the present application is only for description and does not represent the advantages and disadvantages of the embodiment. It should be noted that, in this article, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0143] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0144] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0145] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0146] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0147] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0148] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A traffic complaint processing method based on text mining, characterized in that: Applied to traffic complaint processing equipment, the method comprises: Determine an initial text description vector of a traffic complaint text to be mined, input the initial text description vector of the traffic complaint text to be mined into a text keyword recognition network, and obtain one or more target recognition paragraphs and their target reasoning results; wherein each of the reasoning results is used to indicate whether the target recognition paragraph corresponding to the reasoning of the text keyword recognition network includes a set type of traffic-related entity; Determining whether the target recognition section includes a set type of traffic-related entity through the target reasoning result of each target recognition section; The text keyword recognition network includes the following operations during debugging: Inputting the initial text description vector of each traffic complaint text sample into the basic text network, obtaining target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample, each of which indicates whether the target recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity; The total debugging error is determined by the target reasoning results and actual entity information of each target recognition paragraph, the basic text network is debugged by each traffic complaint text sample until the total debugging error meets the set requirements, and the debugging is stopped, and the text network when the debugging is stopped is determined as a text keyword recognition network, and each actual entity information is used to indicate whether the corresponding target recognition paragraph includes a set type of traffic-related entity; wherein the process of determining each target reasoning result by the basic text network includes: Through the initial text description vector of each traffic complaint text sample, one or more initial recognition paragraphs of the traffic complaint text sample are repeatedly adjusted for a set number of rounds to obtain the corresponding target recognition paragraph, wherein each round of optimization includes: adjusting each first recognition paragraph through the initial text description vector of the traffic complaint text sample to obtain a second recognition paragraph and a corresponding initial reasoning result, each initial reasoning result indicates whether the second recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity, the first recognition paragraph corresponding to each round of optimization is the second recognition paragraph obtained by the previous round of adjustment, the first recognition paragraph corresponding to the first adjustment is each of the initial recognition paragraphs, and the second recognition paragraph obtained by the last round of adjustment is the target recognition paragraph of the traffic complaint text sample; The target reasoning results of each target recognition paragraph of the traffic complaint text sample are determined by the initial reasoning results of the second recognition paragraph obtained through each round of optimization.

2. The method according to claim 1, characterized in that For each of the traffic complaint text samples, determining one or more initial recognition paragraphs of the traffic complaint text sample by using the initial text description vector of the traffic complaint text sample includes: Determine each candidate identification paragraph of the traffic complaint text sample by using the initial text description vector of the traffic complaint text sample; For each of the candidate recognition paragraphs, determine a first overlap ratio between the candidate recognition paragraph and the text paragraph where each traffic-related entity in the traffic complaint text sample is located, and if the first overlap ratio is greater than a first preset overlap ratio, or if each of the first overlap ratios is less than a second preset overlap ratio, determine the candidate recognition paragraph as an initial recognition paragraph of the traffic complaint text sample, and the first preset overlap ratio is greater than the second preset overlap ratio; For each of the traffic complaint text samples, adjusting each first recognized paragraph using the initial text description vector of the traffic complaint text sample to obtain a second recognized paragraph and a corresponding initial reasoning result includes: Projecting the spatial distribution information of each first identified paragraph of the traffic complaint text sample onto the initial text description vector of the traffic complaint text sample to obtain an integrated text description vector; Determine the second recognized paragraphs and the corresponding initial reasoning results corresponding to each first recognized paragraph of the traffic complaint text sample by using the integrated text description vector; For each of the traffic complaint text samples, the initial reasoning result of the second recognition paragraph obtained through each round of optimization is used to determine the target reasoning result of each target recognition paragraph of the traffic complaint text sample, including: For each initial recognition paragraph of the traffic complaint text sample, the integrated reasoning result corresponding to the initial recognition paragraph is determined by the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by each round of optimization except the last round of adjustment, and the target reasoning result of the target recognition paragraph corresponding to the initial recognition paragraph is determined by the integrated reasoning result and the initial reasoning result of the second recognition paragraph corresponding to the initial recognition paragraph obtained by the last round of adjustment.

3. The method according to claim 1, characterized in that The determining of the total debugging error through the target reasoning results and the actual entity information of each target recognition segment includes: Determine a first debugging error through the target reasoning result and actual entity information of each target recognition segment; Determine the text paragraph where each traffic-related entity in each of the traffic complaint text samples is located, and determine the second debugging error through the text paragraph where each traffic-related entity in each of the traffic complaint text samples is located and each of the target recognition paragraphs; A total debugging error is determined by the first debugging error and the second debugging error.

4. The method according to claim 3, characterized in that The determining of the first debugging error by using the target reasoning result and the actual entity information of each target recognition segment includes: Determine a first component error through the target reasoning result and the actual entity information of each target recognition segment, and determine a first debugging error through one or more of a second component error or a third component error and the first component error; Determining the second component error includes: Determine the actual entity information of each second recognition paragraph obtained in each round of optimization; Determine the second composition error through the initial reasoning results and actual entity information of each second recognition segment obtained in each round of optimization; Determining the third component error includes: For each of the initially recognized paragraphs, determining an initial reasoning result of the initially recognized paragraph through the initially recognized paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located; The actual entity information of each of the initially recognized paragraphs is determined, and the third composition error is determined through the initial reasoning results and the actual entity information of each of the initially recognized paragraphs.

5. The method according to claim 4, characterized in that For each of the initially recognized paragraphs, determining the initial reasoning result of the initially recognized paragraph by using the initially recognized paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located includes: If a first overlap rate between the initial recognition paragraph and a text paragraph in which one or more traffic-related entities in a corresponding traffic complaint text sample are located is greater than a first preset overlap rate, it is determined that the text network infers that the initial recognition paragraph includes a set type of traffic-related entity; If the first overlap rate between the initial recognition paragraph and the text paragraph where each traffic-related entity in the corresponding traffic complaint text sample is located is less than the second preset overlap rate, it is determined that the text network infers that the initial recognition paragraph does not include a set type of traffic-related entity.

6. The method according to claim 3, characterized in that: The determining the second debugging error by using the text paragraph where each traffic-related entity in each traffic complaint text sample is located and each target recognition paragraph comprises: Determine a fourth component error through each target identification paragraph and the text paragraph where each traffic-related entity is located in each traffic complaint text sample, and determine a second debugging error through one or more of the fifth component error or the sixth component error and the fourth component error; The fifth component error is determined by the second recognition paragraphs obtained from each round of optimization except the last round of adjustment and the text paragraph where each traffic-related entity is located in each traffic complaint text sample, and the sixth component error is determined by the initial recognition paragraphs and the text paragraph where each traffic-related entity is located in each traffic complaint text sample.

7. The method according to claim 1, characterized in that For each of the target recognition sections, determining whether the target recognition section includes a set type of traffic-related entity based on the target reasoning result of the target recognition section includes: Determine a target entity description vector of a traffic-related entity corresponding to the target recognition segment; Determine a commonality measurement result between the target entity description vector and a preset entity description vector of each set type of traffic-related entity in a description vector library, wherein each of the preset entity description vectors and the target entity description vector is determined by a pre-debugged traffic entity processing operator; Whether the target recognition section includes a set type of traffic-related entity is determined by using the maximum commonality measurement result and the target reasoning result of the target recognition section.

8. The method according to claim 7, characterized in that The determining whether the target identification section includes a set type of traffic-related entity through the maximum commonality measurement result and the target reasoning result of the target identification section includes: If it is determined through the target reasoning result of the target recognition paragraph that the text keyword recognition network infers that the target recognition paragraph does not include a set type of traffic-related entity, and the maximum commonality measurement result is less than the first preset commonality measurement result, it is determined that the target recognition paragraph actually does not include a set type of traffic-related entity; If it is determined through the target inference result of the target recognition paragraph that the text keyword recognition network infers that the target recognition paragraph includes a set type of traffic-related entities, and the maximum commonality measurement result is greater than the second preset commonality measurement result, then it is determined that the target recognition paragraph actually includes the set type of traffic-related entities, and the first preset commonality measurement result is not greater than the second preset commonality measurement result.

9. A traffic complaint handling device, characterized in that: include: A text feature mining module is used to determine the initial text description vector of the traffic complaint text to be mined; A text paragraph reasoning module is used to input the initial text description vector of the traffic complaint text to be mined into a text keyword recognition network to obtain one or more target recognition paragraphs and their target reasoning results; wherein each of the reasoning results is used to indicate whether the target recognition paragraph corresponding to the text keyword recognition network reasoning includes a set type of traffic-related entity; A related entity recognition module, used for determining whether the target recognition section includes a set type of traffic related entity through the target reasoning result of each target recognition section; The recognition network debugging module is used to debug the text keyword recognition network according to the following operations: Inputting the initial text description vector of each traffic complaint text sample into the basic text network, obtaining target reasoning results of one or more target recognition paragraphs of the traffic complaint text sample, each of which indicates whether the target recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity; The total debugging error is determined by the target reasoning results and actual entity information of each target recognition paragraph, the basic text network is debugged by each traffic complaint text sample until the total debugging error meets the set requirements, and the debugging is stopped, and the text network when the debugging is stopped is determined as a text keyword recognition network, and each actual entity information is used to indicate whether the corresponding target recognition paragraph includes a set type of traffic-related entity; wherein the process of determining each target reasoning result by the basic text network includes: Through the initial text description vector of each traffic complaint text sample, one or more initial recognition paragraphs of the traffic complaint text sample are repeatedly adjusted for a set number of rounds to obtain the corresponding target recognition paragraph, wherein each round of optimization includes: adjusting each first recognition paragraph through the initial text description vector of the traffic complaint text sample to obtain a second recognition paragraph and a corresponding initial reasoning result, each initial reasoning result indicates whether the second recognition paragraph corresponding to the text network reasoning includes a set type of traffic-related entity, the first recognition paragraph corresponding to each round of optimization is the second recognition paragraph obtained by the previous round of adjustment, the first recognition paragraph corresponding to the first adjustment is each of the initial recognition paragraphs, and the second recognition paragraph obtained by the last round of adjustment is the target recognition paragraph of the traffic complaint text sample; The target reasoning results of each target recognition paragraph of the traffic complaint text sample are determined by the initial reasoning results of the second recognition paragraph obtained through each round of optimization.

10. A traffic complaint handling device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the steps in the method according to any one of claims 1 to 8 are implemented.

Images