A corpus-based classification method, device, electronic device and storage medium

Abstract

The disclosed embodiments provide a corpus-based classification method, device, electronic device, and storage medium, the method comprising: determining a classification annotation information set according to a target corpus; determining a sentence vector sequence set corresponding to a sentence in the target corpus; inputting the sentence vector sequence set into a trained binary tree classification model to determine the first classification annotation information corresponding to each sentence in the target corpus; inputting the sentence vector sequence set into a trained SVM classification model to determine the second classification annotation information corresponding to each sentence in the target corpus; inputting the sentence vector sequence set into a trained deep learning classification model to determine the third classification annotation information corresponding to each sentence in the target corpus; based on the classification annotation information corresponding to each sentence in the target corpus, determining the failure mode information corresponding to each sentence. The disclosed embodiments can perform fine classification based on corpus while saving computing resources.

Description

Technical Field

[0001] The present invention relates to the field of computer software technology, and in particular to a corpus-based classification method, device, electronic equipment and storage medium. Background Art

[0002] With the deep integration of industrialization and informatization, enterprises have access to more and more data resources. Therefore, it is necessary to study how to better utilize data resources to fully utilize data resources, improve product quality, and improve enterprise manufacturing efficiency. In the process of industrial research and development and production, it is often necessary to monitor the failure of the products sold, so there is a widespread demand for quality problem classification based on corpus. The causes of quality problems are complex. Different quality problems have different influencing factors, different scenarios, and different objects of quality problems.

[0003] The method of extracting and classifying quality problems based on corpus can directly mine unstructured data on the Internet, use natural language processing tools to extract online comments and other information, and classify the text, and finally convert the unstructured data into structured data, and accumulate and count the frequency of each type of problem for each product model and store it in the database. However, the current classification method cannot classify quality problems in a sufficiently detailed manner, resulting in the inability of enterprises to fully utilize the massive corpus resources and to better improve the quality of their products. Summary of the invention

[0004] In view of the defects in the prior art, the embodiments of the present disclosure provide a corpus-based classification method, which can perform fine classification based on the corpus while saving computing resources.

[0005] The embodiment of the present disclosure provides a corpus-based classification method, the method comprising: determining a classification annotation information set according to a target corpus; the classification annotation information set comprises a first classification annotation information set, a second classification annotation information set and a third classification annotation information set; determining a sentence vector sequence set corresponding to sentences in the target corpus; inputting the sentence vector sequence set into a trained binary tree classification model to determine the first classification annotation information corresponding to each sentence in the target corpus; inputting the sentence vector sequence set into a trained SVM classification model to determine the second classification annotation information corresponding to each sentence in the target corpus; inputting the sentence vector sequence set into a trained deep learning classification model to determine the third classification annotation information corresponding to each sentence in the target corpus; based on the first classification annotation information, the second classification annotation information and the third classification flag information corresponding to each sentence in the target corpus, determining the failure mode information corresponding to each sentence.

[0006] In an optional embodiment, the first classification labeling information set includes structural information of the car and / or system information of the car, the second classification labeling information set includes performance evaluation index information of the car, and the third classification labeling information set includes scene information of the car.

[0007] In an optional embodiment, the trained deep learning classification model is a trained Bert classification model.

[0008] In an optional embodiment, determining a classification label information set according to the target corpus includes: extracting key feature information in the target corpus based on a feature extraction network; screening and classifying the key feature information to determine the classification label information set.

[0009] In an optional embodiment, before determining the classification annotation information set based on the target corpus, the method also includes: obtaining the original corpus; preprocessing the original corpus to obtain the target corpus; the preprocessing includes at least one of extracting useful information, word segmentation, removing stop words, sentence segmentation, and structuring.

[0010] In an optional embodiment, the classification annotation information set also includes a fourth classification annotation information set, and the fourth classification annotation information set includes sentiment tendency feature information; after inputting the sentence vector sequence set into a trained deep learning classification model to determine the third classification annotation information corresponding to each sentence in the target corpus, the method also includes: inputting the word vector sequence set into a trained deep learning sentiment classification model to determine the fourth classification annotation information corresponding to each sentence in the target corpus.

[0011] Accordingly, an embodiment of the present disclosure provides a corpus-based classification device, the device comprising:

[0012] A labeling module, used to determine a classification labeling information set according to the target corpus; the classification labeling information set includes a first classification labeling information set, a second classification labeling information set and a third classification labeling information set;

[0013] A sentence vector determination module is used to determine a set of sentence vector sequences corresponding to sentences in the target corpus;

[0014] The first classification module is used to input the sentence vector sequence set into the trained binary tree classification model to determine the first classification labeling information corresponding to each sentence in the target corpus;

[0015] The second classification module is used to input the sentence vector sequence set into the trained SVM classification model to determine the second classification labeling information corresponding to each sentence in the target corpus;

[0016] The third classification module is used to input the sentence vector sequence set into the trained deep learning classification model to determine the third classification labeling information corresponding to each sentence in the target corpus;

[0017] The failure mode determination module is used to determine the failure mode information corresponding to each sentence based on the first classification annotation information, the second classification annotation information and the third classification mark information corresponding to each sentence in the target corpus.

[0018] In an optional embodiment, the first classification labeling information set includes structural information of the car and / or system information of the car, the second classification labeling information set includes performance evaluation index information of the car, and the third classification labeling information set includes scene information of the car.

[0019] In an optional embodiment, the trained deep learning classification model is a trained Bert classification model.

[0020] In an optional embodiment, the annotation module is used to extract key feature information in the target corpus based on a feature extraction network; screen and classify the key feature information to determine a classification annotation information set.

[0021] In an optional embodiment, the corpus-based classification device also includes an acquisition module for acquiring original corpus; preprocessing the original corpus to obtain target corpus; the preprocessing includes at least one of extracting useful information, word segmentation, removing stop words, sentence segmentation, and structuring.

[0022] In an optional embodiment, the classification annotation information set also includes a fourth classification annotation information set, and the fourth classification annotation information set includes sentiment tendency feature information; the corpus-based classification device also includes a fourth classification module, which is used to input the sentence vector sequence set into the trained deep learning classification model to determine the third classification annotation information corresponding to each sentence in the target corpus, and then input the word vector sequence set into the trained deep learning sentiment classification model to determine the fourth classification annotation information corresponding to each sentence in the target corpus.

[0023] Accordingly, an embodiment of the present disclosure provides an electronic device, characterized in that the electronic device includes a processor and a memory, the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the above-mentioned corpus-based classification method.

[0024] Accordingly, an embodiment of the present disclosure provides a computer-readable storage medium, characterized in that at least one instruction, at least one program, code set or instruction set is stored in the storage medium, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the above-mentioned corpus-based classification method.

[0025] Accordingly, an embodiment of the present disclosure provides a computer program product, characterized in that the computer program product includes a computer program, the computer program is stored in a readable storage medium, and at least one processor of a computer device reads and executes the computer program from the readable storage medium, so that the computer device executes the above-mentioned corpus-based classification method.

[0026] The embodiments of the present application have the following beneficial effects:

[0027] (1) Multi-level classification annotation information includes annotation information of various dimensions, which can more accurately classify the content in the corpus information;

[0028] (2) Perform multi-level classification through different models to save computing resources to the greatest extent while meeting classification requirements;

[0029] (3) It can be used to analyze quality issues and generate failure mode analysis reports to help companies deal with and respond to quality issues in a targeted manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 It is a schematic diagram of an application scenario of a corpus-based classification method provided in an embodiment of the present application;

[0032] Figure 2 It is a flowchart of a corpus-based classification method provided in an embodiment of the present application;

[0033] Figure 3 The embodiment of the present application provides Figure 2 The flowchart of step S201 is shown;

[0034] Figure 4 It is a flowchart of a corpus-based classification method provided in an embodiment of the present application;

[0035] Figure 5It is a schematic diagram of an analysis report of a corpus-based classification method provided in an embodiment of the present application;

[0036] Figure 6 It is a structural schematic diagram of a corpus-based classification device provided in an embodiment of the present application;

[0037] Figure 7 This is a hardware structure block diagram of a server for a corpus-based classification method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiment is only one embodiment of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0039] The "embodiment" referred to herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present application. In the description of the embodiments of the present application, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "left", "right", "top", "bottom", etc. is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device / system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present application. The terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may include one or more of the features explicitly or implicitly. Moreover, the terms "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" and "having" / "for" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system / apparatus, product or apparatus comprising a series of steps or units / modules is not necessarily limited to those steps or units / modules explicitly listed, but may include other steps or units / modules not explicitly listed or inherent to such process, method, product or apparatus.

[0040] The following is a specific embodiment of a corpus-based classification method provided by this application. Figure 1 , Figure 1Schematic diagram of an application scenario of a corpus-based classification method provided in an embodiment of the present application. Figure 1 As shown, it includes a server 101 and a terminal 102. Optionally, the server 101 and the terminal 102 may be connected via a wireless link or a wired link, which is not limited in the present disclosure.

[0041] In an optional embodiment, the server 101 can be used to classify the acquired target corpus to obtain the failure mode information corresponding to each sentence. Specifically, the server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the operating system running on the server 01 may include but is not limited to IOS, Linux, Windows, Unix, Android systems, etc.

[0042] In an optional embodiment, the terminal 102 can assist the server 101 in obtaining the failure mode information corresponding to each sentence. The terminal can be a provider of the target corpus, and is used to send the target corpus to the server 101. Optionally, after the server 101 completes the classification, the classified target corpus and the failure mode information corresponding to each sentence in the target corpus can be sent back to the terminal 102 for display on the terminal 102. Specifically, the terminal 102 can include but is not limited to electronic devices such as smart phones, desktop computers, tablet computers, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, smart wearable devices, etc. Optionally, the operating system running on the electronic device can include but is not limited to Android system, IOS system, Linux, Windows, etc.

[0043] In addition, it should be noted that Figure 1 What is shown is only one application environment of the corpus-based classification method provided by the present disclosure. In practical applications, other application environments may also be included. For example, classification is performed based on the acquired target corpus to obtain failure mode information corresponding to each sentence, which may also be implemented on the terminal 102.

[0044] Figure 2It is a flowchart of a corpus-based classification method provided in an embodiment of the present application. This specification provides method operation steps as shown in the embodiment or flowchart, but more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiment is only one of many execution orders and does not represent the only execution order. In actual execution, the method sequence shown in the embodiment or the drawings may be executed or executed in parallel (for example, in a parallel processor or multi-threaded processing environment). Specifically, Figure 2 As shown, the method includes:

[0045] S201: Determine a classification annotation information set according to the target corpus.

[0046] In an optional implementation, the classification labeling information set includes a first classification labeling information set, a second classification labeling information set, and a third classification labeling information set.

[0047] Figure 3 The embodiment of the present application provides Figure 2 The flowchart of step S201 is shown.

[0048] Combine the following Figure 3 The illustrated process further describes the method of step S201. Figure 3 As shown, the illustrated method includes:

[0049] S301: Extract key feature information from the target corpus based on the feature extraction network.

[0050] In an optional implementation, the feature extraction network may be a dependency syntax analysis network. Optionally, based on the dependency syntax analysis network, key feature information such as core components, agents, patients, qualifiers, descriptive aspects, and degree words of any sentence in the target corpus may be extracted.

[0051] According to some embodiments, the sentence input into the dependency parsing network may be "the engine makes a loud noise when accelerating", then the extracted core component may be "acceleration", the agent may be "engine", the descriptive body may be "big", and the qualifier may be "noise".

[0052] S302: Screen and classify the key feature information to determine a classification annotation information set.

[0053] Optionally, the extracted key feature information can be screened by manual screening to remove useless key feature information and merge similar key feature information. Optionally, the key feature information can be classified into multiple levels based on the quality evaluation system to determine the classification annotation information set.

[0054] In an optional implementation, the first classification labeling information set may include the structural information of the car or the system information of the car. Optionally, the first classification labeling information in the first classification labeling information set may be the five major systems of the car. Optionally, the first classification labeling information may be any one of the chassis suspension and the engine. It should be noted that the first classification labeling information is not limited to the content in the above embodiments. In other embodiments, the first classification labeling information may also be other optional content.

[0055] In an optional implementation, the second classification labeling information set may include performance evaluation index information of the automobile. Optionally, the second classification labeling information in the second classification labeling information set may correspond to the first classification labeling information, and a single first classification labeling information may correspond to multiple second classification labeling information.

[0056] According to some embodiments, the second classification labeling information may be handling performance, comfort, and the second classification labeling information may correspond to the same first classification labeling information. Optionally, the first classification labeling information may be chassis suspension.

[0057] According to some embodiments, the second classification labeling information may be noise, fuel consumption, and these second classification labeling information may correspond to the same first classification labeling information. Optionally, the first classification labeling information may be an engine.

[0058] It should be noted that the second classification labeling information is not limited to the content in the above embodiment. In other embodiments, the second classification labeling information may also be other optional content.

[0059] In an optional implementation, the third classification labeling information may include scene information of the car. Optionally, the third classification labeling information in the third classification labeling information set may correspond to the second classification labeling information, and a single second classification labeling information may correspond to a plurality of third classification labeling information.

[0060] According to some embodiments, the third classification labeling information may be on a highway, on a rural road, or on a bad road, and these third classification labeling information may correspond to the same second classification labeling information. Optionally, the second classification labeling information may be handling performance.

[0061] According to some embodiments, the third classification labeling information may be during acceleration, during starting, or during gear shifting, and these third classification labeling information may correspond to the same second classification labeling information. Optionally, the second classification labeling information may be noise.

[0062] According to some embodiments, the third classification labeling information may be on a speed bump, in an urban area, and these third classification labeling information may correspond to the same second classification labeling information. Optionally, the second classification labeling information may be handling performance.

[0063] According to some embodiments, the third classification labeling information may be on a highway, at idle speed, in an urban area, in winter, and in summer, and these third classification labeling information may correspond to the same second classification labeling information set. Optionally, the second classification labeling information may be fuel consumption.

[0064] It should be noted that the third classification labeling information is not limited to the content in the above embodiments. In other embodiments, the third classification labeling information may also be other optional content.

[0065] In an optional implementation, the classification annotation information set may further include a fourth classification annotation information set, and the fourth classification annotation information set may include sentiment tendency feature information. Optionally, the fourth classification annotation information in the fourth classification annotation information set may be positive, negative, or neutral. It should be noted that the fourth classification annotation information is not limited to the content in the above embodiment, and in other embodiments, the fourth classification annotation information may also be other optional content.

[0066] In an optional implementation, after step S201, a trained binary tree classification model, a trained SVM classification model, and a trained deep learning classification model can be trained based on the first classification labeling information set, the second classification labeling information set, and the third classification labeling information set, respectively.

[0067] According to some embodiments, before training the binary tree classification model, a binary tree training corpus may be obtained to establish an initial binary tree classification model. Optionally, the binary tree training corpus may be obtained from a target corpus.

[0068] According to some embodiments, an initial binary tree classification model is trained based on the binary tree training corpus and the first classification annotation information set to obtain a trained binary tree classification model.

[0069] According to some embodiments, before training the SVM classification model, SVM training corpus may be obtained to establish an initial SVM classification model. Optionally, the SVM training corpus may be obtained from the target corpus.

[0070] According to some embodiments, an initial SVM classification model is trained based on the SVM training corpus and the second classification annotation information set to obtain a trained binary tree classification model. Optionally, the initial SVM classification model can be determined as the current SVM classification model. Based on the current SVM classification model, the second classification annotation information corresponding to each sentence in the training SVM corpus is determined; based on the second classification annotation information corresponding to each sentence in the training SVM corpus, an optimization target is determined; based on the optimization target, the current SVM classification model is updated, and the updated SVM classification model is re-determined as the current SVM classification model; if it is determined that there is no optimization target or it is determined that the second classification annotation information corresponding to each sentence in the training SVM corpus meets the preset classification requirements, the current SVM classification model is determined as the trained SVM classification model, otherwise the above steps are repeated.

[0071] According to some embodiments, before training the deep learning classification model, deep learning training corpus may be obtained to establish an initial deep learning classification model. Optionally, the deep learning training corpus may be obtained from the target corpus.

[0072] According to some embodiments, an initial deep learning model is trained based on the deep learning training corpus and the third classification annotation information set to obtain a trained deep learning model. Optionally, the initial deep learning classification model can be determined as the current deep learning classification model. Based on the current deep learning classification model, the third classification annotation information corresponding to each sentence in the training deep learning corpus is determined; based on the third classification annotation information corresponding to each sentence in the training deep learning corpus, an optimization target is determined; based on the optimization target, the current deep learning classification model is updated, and the updated deep learning classification model is re-determined as the current deep learning classification model; if it is determined that there is no optimization target or it is determined that the second classification annotation information corresponding to each sentence in the training deep learning corpus meets the preset classification requirements, the current deep learning classification model is determined as the trained deep learning classification model, otherwise the above steps are repeated.

[0073] S202: Determine a set of sentence vector sequences corresponding to sentences in the target corpus.

[0074] Optionally, the sentence vector sequence in the sentence vector sequence set may be a set of word vector sequences corresponding to all words contained in the sentence.

[0075] Optionally, a weighted average operation may be performed on each dimension of the word vector sequence in the word vector sequence set to obtain a sentence vector sequence set corresponding to the sentences in the target corpus information.

[0076] In the embodiment of the present application, the models used for the classification method include a binary tree classification model, a SVM classification model and a deep learning classification model. Optionally, the deep learning classification model can be a trained Bert classification model.

[0077] S203: Input the sentence vector sequence set into the trained binary tree classification model to determine the first classification labeling information corresponding to each sentence in the target corpus.

[0078] S204: Input the sentence vector sequence set into the trained SVM classification model to determine the second classification labeling information corresponding to each sentence in the target corpus.

[0079] S205: Input the sentence vector sequence set into the trained deep learning classification model to determine the third classification labeling information corresponding to each sentence in the target corpus.

[0080] In an optional implementation, in a binary tree classification model, an ordered sequence of binary tree nodes can be determined based on a first classification labeling information set. Optionally, based on the ordered sequence of binary tree nodes, the word vector sequence and / or sentence vector sequence corresponding to each sentence in the word vector sequence set can be traversed to determine the first classification labeling information corresponding to each sentence. Performing the first level of classification through a binary tree classification model is equivalent to performing a binary search on an ordered sequence of binary tree nodes. This model has a low structural complexity and can meet classification requirements without wasting computing power.

[0081] In an optional implementation, the deep learning model may be a Bert model. Since the Bert model has been pre-trained with massive data, the expected classification effect can be achieved with only a small amount of training. Therefore, the Bert model is used for complex third-level classification of corpus, which can save computing power while meeting classification requirements.

[0082] According to some embodiments, the sentence vector sequence set can be input into a trained SVM classification model, and based on the first classification labeling information corresponding to each sentence, the second classification labeling information corresponding to each sentence in the target corpus can be determined. Based on the first classification labeling information of each sentence, multiple second classification labeling information corresponding to the first classification labeling information can be determined, thereby narrowing the optional labeling range in the secondary classification of each sentence, thereby reducing the amount of calculation.

[0083] According to some embodiments, the sentence vector sequence set can be input into the trained Bert classification model, and the third classification labeling information corresponding to each sentence in the target corpus can be determined based on the second classification labeling information corresponding to each sentence. Based on the second classification labeling information of each sentence, multiple third classification labeling information corresponding to the second classification labeling information can be determined, thereby narrowing the optional labeling range in the three-level classification of each sentence, thereby reducing the amount of calculation.

[0084] In an optional implementation, the sentence "The chassis is too hard when driving over a speed bump" is classified. The sentence vector corresponding to the sentence can be input into a trained binary tree classification model to determine that the first classification labeling information corresponding to the sentence is "chassis"; the sentence vector corresponding to the sentence is input into a trained SVM classification model to determine that the second classification labeling information corresponding to the sentence is "comfort"; the sentence vector corresponding to the sentence is input into a trained deep learning classification model to determine that the third classification labeling information corresponding to the sentence is "on the speed bump".

[0085] In an optional implementation, the sentence "The fuel consumption is very high on the highway" is classified. The sentence vector corresponding to the sentence can be input into a trained binary tree classification model to determine that the first classification labeling information corresponding to the sentence is "fuel system"; the sentence vector corresponding to the sentence is input into a trained SVM classification model to determine that the second classification labeling information corresponding to the sentence is "fuel consumption"; the sentence vector corresponding to the sentence is input into a trained deep learning classification model to determine that the third classification labeling information corresponding to the sentence is "on the highway".

[0086] Optionally, the classification annotation information set may further include a fourth classification annotation information set, and the fourth classification annotation information set may include sentiment tendency features.

[0087] In an optional embodiment, after the word vector sequence set is input into a trained deep learning classification model to determine the third category labeling information corresponding to each sentence in the target corpus, the sentence vector sequence set can be input into a trained deep learning sentiment classification model to determine the fourth category labeling information corresponding to each sentence in the target corpus.

[0088] Optionally, a deep learning sentiment classification model may be trained according to the fourth classification annotation information set to obtain a trained deep learning classification model. Optionally, the deep learning sentiment classification model may be a Bert model.

[0089] According to the above sentence "the chassis is too hard when passing over a speed bump", the fourth category labeling information corresponding to the sentence may be "too hard" or "negative". According to the above sentence "the fuel consumption is very high on the highway", the fourth category labeling information corresponding to the sentence may be "very high" or "negative".

[0090] Through the level-by-level classification, not only the location information of the automobile parts or systems mentioned in the corpus is marked, but also the failure mode information and scenario information of the automobile is marked. The key information in the corpus is marked more comprehensively, which is convenient for analyzing and managing product quality issues. Level-by-level classification uses a model suitable for each level, and the relationship between each level of label and each level of classification can be progressive, which can save the computing resources required for a single classification model.

[0091] S206: Determine failure mode information corresponding to each sentence based on the first classification annotation information, the second classification annotation information, and the third classification mark information corresponding to each sentence in the target corpus.

[0092] In an optional implementation, based on the sentence "the chassis is too hard when driving over a speed bump", the first category annotation information corresponding to the sentence is "chassis", the second category annotation information is "comfort", and the third category annotation information is "on the speed bump", it can be determined that the failure mode information corresponding to the sentence is "chassis comfort on the speed bump".

[0093] In an optional implementation, based on the statement "the fuel consumption is very high on the highway", the first category labeling information corresponding to the statement is "fuel system", the second category labeling information is "fuel consumption", and the third category labeling information is "on the highway", it can be determined that the failure mode information corresponding to the statement is "fuel system consumes fuel on the highway".

[0094] In an optional implementation, based on the sentence "the chassis is too hard when driving over a speed bump", the first category annotation information corresponding to the sentence is "chassis", the second category annotation information is "comfort", the third category annotation information is "on the speed bump", and the fourth category annotation information is "too hard", it can be determined that the failure mode information corresponding to the sentence is "the chassis is too hard for comfort on the speed bump".

[0095] In an optional implementation, based on the statement "the fuel consumption is very high on the highway", the first category labeling information corresponding to the statement is "fuel system", the second category labeling information is "fuel consumption", the third category labeling information is "on the highway", and the fourth category labeling information is "very high", it can be determined that the failure mode information corresponding to the statement is "the fuel system consumes very high fuel on the highway".

[0096] Figure 4 FIG. 1 is a flow chart of a corpus-based classification method provided in an embodiment of the present application. Figure 4 As shown, the illustrated method includes:

[0097] S401: Obtain original corpus.

[0098] In an optional implementation, the content of the social media website is collected by an information collection tool, and the content of the social media website is extracted by an information analysis tool to obtain the original corpus. Optionally, the original corpus may include social media review content of a single car model.

[0099] S402: Preprocess the original corpus to obtain the target corpus.

[0100] Optionally, the preprocessing includes at least one of extracting useful information, word segmentation, removing stop words, sentence segmentation, and structuring. In an optional implementation, the original corpus information can be input into a trained CBOW or Skip-gram model to obtain a word vector set in the original corpus information. Optionally, after removing stop words and high-frequency words, the target corpus information and the word vector sequence set of the target corpus information can be determined. Optionally, a weighted average operation can be performed on each dimension of the word vector in the word vector set to obtain a sentence vector sequence set corresponding to the sentence in the target corpus information.

[0101] S403: Determine a classification annotation information set according to the target corpus.

[0102] S404: Determine a set of sentence vector sequences corresponding to sentences in the target corpus.

[0103] S405: Input the sentence vector sequence set into the trained binary tree classification model to determine the first classification labeling information corresponding to each sentence in the target corpus.

[0104] S406: Input the sentence vector sequence set into the trained SVM classification model to determine the second classification labeling information corresponding to each sentence in the target corpus.

[0105] S407: Input the sentence vector sequence set into the trained deep learning classification model to determine the third classification labeling information corresponding to each sentence in the target corpus.

[0106] S408: Determine failure mode information corresponding to each sentence based on the first classification annotation information, the second classification annotation information, and the third classification mark information corresponding to each sentence in the target corpus.

[0107] For the specific contents of step S403 to step S408, reference may be made to the description of the above-mentioned step S201 to step S206, which will not be repeated here.

[0108] S409: Determine a failure mode analysis report based on the failure mode information corresponding to each sentence in the target corpus.

[0109] Figure 5 This is a schematic diagram of an analysis report of a corpus-based classification method provided in an embodiment of the present application. Figure 5 The method of step S409 is further described.

[0110] In an optional implementation, the number of statements corresponding to each failure mode information may be counted and tabulated to obtain a failure mode analysis report. Optionally, the failure mode analysis report may be an analysis report for a specific vehicle model. Figure 5 In the figure, the horizontal axis of the number axis can be a set of failure mode information containing failure mode information; the vertical axis of the number axis can represent the number of sentences corresponding to each failure mode information in the target corpus, and the larger the number, the higher the frequency of occurrence of the failure mode. Optionally, an overall sentiment analysis report can also be generated to count the overall sentiment tendency reflected in the corpus.

[0111] The embodiment of the present application can make full use of the information related to quality issues in the corpus by determining the failure mode analysis report, and help enterprises to handle and respond to quality issues in a directionally.

[0112] The present application also provides a corpus-based classification device. Figure 6 is a schematic diagram of a corpus-based classification device 600 provided in an embodiment of the present application, such as Figure 6 As shown, the device comprises:

[0113] The annotation module 601 can be used to determine a classification annotation information set according to the target corpus; the classification annotation information set includes a first classification annotation information set, a second classification annotation information set and a third classification annotation information set;

[0114] A sentence vector determination module 602 is used to determine a set of sentence vector sequences corresponding to sentences in the target corpus;

[0115] The first classification module 603 may be used to input the sentence vector sequence set into the trained binary tree classification model to determine the first classification labeling information corresponding to each sentence in the target corpus;

[0116] The second classification module 604 may be used to input the sentence vector sequence set into the trained SVM classification model to determine the second classification labeling information corresponding to each sentence in the target corpus;

[0117] The third classification module 605 can be used to input the sentence vector sequence set into the trained deep learning classification model to determine the third classification labeling information corresponding to each sentence in the target corpus;

[0118] The failure mode determination module 606 may be used to determine the failure mode information corresponding to each sentence based on the first classification labeling information, the second classification labeling information and the third classification flag information corresponding to each sentence in the target corpus.

[0119] In an optional embodiment, the first classification labeling information set may include structural information of the car and / or system information of the car, the second classification labeling information set may include performance evaluation index information of the car, and the third classification labeling information set includes scene information of the car.

[0120] In an optional embodiment, the trained deep learning classification model may be a trained Bert classification model.

[0121] In an optional embodiment, the annotation module 601 can also be used to extract key feature information in the target corpus based on a feature extraction network; screen and classify the key feature information to determine a classification annotation information set.

[0122] In an optional embodiment, the corpus-based classification device 600 may also include an acquisition module for acquiring original corpus; preprocessing the original corpus to obtain target corpus; the preprocessing may include at least one of extracting useful information, word segmentation, removing stop words, sentence segmentation, and structuring.

[0123] In an optional embodiment, the classification annotation information set may also include a fourth classification annotation information set, and the fourth classification annotation information set includes sentiment tendency feature information; the corpus-based classification device 600 may also include a fourth classification module, which is used to input the sentence vector sequence set into the trained deep learning classification model to determine the third classification annotation information corresponding to each sentence in the target corpus, and then input the word vector sequence set into the trained deep learning sentiment classification model to determine the fourth classification annotation information corresponding to each sentence in the target corpus.

[0124] The device in the embodiment of the present application is based on the same application concept as the method embodiment described above.

[0125] The present application also provides an electronic device, which can be set in a server to store a corpus-based classification method for implementing the method embodiment. The method embodiment provided in the present application can be executed in a computer terminal, a server or a similar computing device. Taking running on a server as an example, Figure 7 is a hardware structure block diagram of a server of a corpus-based classification method provided in an embodiment of the present application. Figure 7As shown, the server 700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 710 (the processor 710 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 730 for storing data, and one or more storage media 720 (such as one or more mass storage devices) for storing application programs 723 or data 722. Among them, the memory 730 and the storage medium 720 can be short-term storage or permanent storage. The program stored in the storage medium 720 may include one or more modules, each of which may include a series of instruction operations on the server. Furthermore, the central processing unit 710 can be configured to communicate with the storage medium 720 and execute a series of instruction operations in the storage medium 720 on the server 700. The server 700 may also include one or more power supplies 760, one or more wired or wireless network interfaces 750, one or more input and output interfaces 740, and / or one or more operating systems 721, such as Windows ServerTM, Mac OSXTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0126] The input / output interface 740 may be used to receive or send data via a network. The specific example of the network may include a wireless network provided by a communication provider of the server 700. In one example, the input / output interface 740 includes a network adapter (Network Interface Controller, NIC), which may be connected to other network devices via a base station so as to communicate with the Internet. In one example, the input / output interface 740 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0127] It can be understood by those skilled in the art that Figure 7 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 7 More or fewer components as shown, or with Figure 7 Different configurations are shown.

[0128] An embodiment of the present application also provides a storage medium, which can be set in a server to store at least one instruction, at least one program, a code set or an instruction set related to the corpus-based classification method in the method embodiment. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the above-mentioned corpus-based classification method.

[0129] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a mobile hard disk, a magnetic disk, or an optical disk.

[0130] In the present invention, unless otherwise clearly specified and limited, the terms "connected", "connection" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be a connection between two elements or an interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0131] It should be noted that the order of the above embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. The above description describes specific embodiments, and other embodiments are also within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in the order of different embodiments and can achieve the expected results. In addition, the processes depicted in the drawings do not necessarily require a specific order or connection order to achieve the desired results. In some embodiments, multi-task parallel processing is also possible or may be advantageous.

[0132] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device / system, since they are based on similarities to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0133] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A corpus-based classification method, characterized in that: include: Determine the classification annotation information set based on the target corpus; The classification labeling information set includes a first classification labeling information set, a second classification labeling information set and a third classification labeling information set that are multi-level classified based on a quality evaluation system; Determine a sentence sequence set in the target corpus and input it into a trained binary tree classification model to determine each corresponding sentence vector sequence set in the target corpus; Labeling the first classification information corresponding to the sentences in the sentence vector; Inputting the sentence vector sequence set into a trained SVM classification model, and determining the second classification annotation information corresponding to each sentence in the target corpus based on the first classification annotation information corresponding to each sentence, wherein a single first classification annotation information corresponds to multiple second classification annotation information; Inputting the sentence vector sequence set into a trained deep learning classification model, determining the third classification annotation information corresponding to each sentence in the target corpus based on the second classification annotation information corresponding to each sentence, wherein a single second classification annotation information corresponds to multiple third classification annotation information, and the first classification annotation information, the second classification annotation information, and the third classification annotation information are obtained by progressive classification, and belong to annotation information of different dimensions, wherein each level of model, each level of label, and each level of classification are in a progressive relationship; Based on the first classification labeling information, the second classification labeling information, and the third classification labeling information corresponding to each sentence in the target corpus, failure mode information corresponding to each sentence is determined.

2. The corpus-based classification method according to claim 1, characterized in that: The first classification labeling information set includes structural information of the car and / or system information of the car, the second classification labeling information set includes performance evaluation index information of the car, and the third classification labeling information set includes scene information of the car.

3. The corpus-based classification method according to claim 1, characterized in that: The trained deep learning classification model is a trained Bert classification model.

4. The corpus-based classification method according to claim 1, characterized in that: The step of determining a classification label information set according to the target corpus includes: Extracting key feature information from the target corpus based on a feature extraction network; The key feature information is screened and classified to determine the classification labeling information set.

5. The corpus-based classification method according to claim 1, characterized in that: Before determining the classification annotation information set according to the target corpus, the method further includes: Get the original corpus; Preprocessing the original corpus to obtain the target corpus; The preprocessing includes at least one of extracting useful information, word segmentation, removing stop words, segmenting sentences, and structuring.

6. The corpus-based classification method according to claim 1, characterized in that: The classification labeling information set further includes a fourth classification labeling information set, and the fourth classification labeling information set includes emotional tendency feature information; After inputting the sentence vector sequence set into a trained deep learning classification model to determine the third classification labeling information corresponding to each sentence in the target corpus, the method further includes: The sentence vector sequence set is input into a trained deep learning sentiment classification model to determine fourth category labeling information corresponding to each sentence in the target corpus.

7. A corpus-based classification device, characterized in that: The device comprises: A labeling module, used to determine a classification labeling information set according to the target corpus; the classification labeling information set includes a first classification labeling information set, a second classification labeling information set and a third classification labeling information set that are multi-level classified based on a quality evaluation system; A sentence vector determination module, used to determine a set of sentence vector sequences corresponding to sentences in the target corpus; A first classification module, used to input the sentence vector sequence set into a trained binary tree classification model to determine first classification labeling information corresponding to each sentence in the target corpus; A second classification module is used to input the sentence vector sequence set into a trained SVM classification model, and determine the second classification annotation information corresponding to each sentence in the target corpus based on the first classification annotation information corresponding to each sentence, where a single first classification annotation information corresponds to multiple second classification annotation information; A third classification module is used to input the sentence vector sequence set into a trained deep learning classification model, and determine the third classification annotation information corresponding to each sentence in the target corpus based on the second classification annotation information corresponding to each sentence, wherein a single second classification annotation information corresponds to multiple third classification annotation information, and the first classification annotation information, the second classification annotation information and the third classification annotation information are obtained by progressive classification, and belong to annotation information of different dimensions, wherein each level of model, each level of label and each level of classification are in a progressive relationship; A failure mode determination module is used to determine the failure mode information corresponding to each sentence in the target corpus based on the first classification annotation information, the second classification annotation information and the third classification annotation information corresponding to each sentence.

8. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the corpus-based classification method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the corpus-based classification method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product includes a computer program, which is stored in a readable storage medium. At least one processor of a computer device reads and executes the computer program from the readable storage medium, so that the computer device executes the corpus-based classification method as described in any one of claims 1 to 6.

Images