Vehicle part affiliation determination method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a vehicle part affiliation determination method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining an appeal text corresponding to a user appeal; inputting the appeal text into a preset rule model; the rule model is used for identifying component affiliation matched with the appeal text and outputting an identification result based on a preset vehicle component and component affiliation mapping dictionary; if the rule model identifies the component affiliation, taking the component affiliation as the vehicle component affiliation corresponding to the user appeal; otherwise, obtaining feature information corresponding to the appeal text; inputting the feature information into a preset classification model, and determining vehicle component affiliation corresponding to the user appeal based on classification probability information output by the classification model; the classification model is used for determining the probability that the feature information belongs to the multiple preset parts and outputting the probability. According to the method, the affiliation of the vehicle component in the user appeal is determined, the consumed time is shorter, and the efficiency and the accuracy are better.

Description

Technical Field

[0001] The present application relates to the field of vehicle technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for determining ownership of vehicle parts. Background Art

[0002] For automobile manufacturers or service providers, they often receive various demands from users regarding vehicles, namely user demands. At this time, it is necessary to find the vehicle parts involved and the corresponding parts ownership from the user demand information to better serve the user, that is, to analyze and identify the ownership of the automobile parts in the user demand. For example, if the user's demand is "there is a problem with the window lifting and lowering when driving at high speed", it is necessary to identify that the part corresponding to the demand belongs to the car body, so as to facilitate subsequent inspection, maintenance and other related work on the car body.

[0003] Therefore, how to accurately determine the ownership of vehicle parts in user demands is a problem that needs to be solved. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for determining the ownership of vehicle parts in response to the above technical problems, so as to accurately determine the ownership of vehicle parts in user requests.

[0005] In a first aspect, the present application provides a method for determining ownership of vehicle parts, comprising:

[0006] Get the demand text corresponding to the user's demand;

[0007] Input the appeal text into a preset rule model; the rule model is used to identify the attribution of the parts matching the appeal text based on a preset mapping dictionary between vehicle parts and part attribution, and output the recognition result;

[0008] When the output representation of the rule model identifies the attribution of the component, the attribution of the component is used as the attribution of the vehicle component corresponding to the user's demand; when the output representation of the rule model does not identify the attribution of the component, the feature information corresponding to the demand text is obtained;

[0009] The feature information is input into a preset classification model, and the vehicle component attribution corresponding to the user's demand is determined based on the classification probability information output by the classification model; the classification model is used to determine and output the probability that the feature information belongs to each component among multiple preset component attributions.

[0010] In one embodiment, obtaining feature information corresponding to the appeal text includes:

[0011] Extract features from the appeal text to obtain the expression form features and semantic features of the appeal text;

[0012] Based on the expression form features, semantic features and their corresponding weights, the general features of the appeal text are obtained;

[0013] The general features, expression features and semantic features are integrated to determine the feature information corresponding to the appeal text.

[0014] In one embodiment, the expression form features of the appeal text include phonetic features and glyph features of the appeal text;

[0015] Based on the expression features, semantic features and their corresponding weights, the general features of the appeal text are obtained, including:

[0016] The weighted average of the phonetic features, glyph features and semantic features is used as the general features of the appeal text;

[0017] By integrating the general features, expression features and semantic features, the feature information corresponding to the appeal text is determined, including:

[0018] Combine the phonetic features, glyph features and semantic features to obtain the detailed features corresponding to the appeal text;

[0019] Combine the general features with the detailed features to obtain the feature information of the appeal text.

[0020] In one embodiment, feature extraction is performed on the appeal text to obtain expression form features and semantic features of the appeal text, including:

[0021] Extract the pinyin features of the appeal text based on the annotated pinyin data;

[0022] Based on the text font, extract the font features of the appeal text;

[0023] Based on the preset semantic analysis model, the semantic features of the appeal text are extracted.

[0024] In one embodiment, the method further comprises:

[0025] Based on multiple vehicle parts and multiple part attributions, construct an initial mapping dictionary between vehicle parts and part attributions;

[0026] Construct a prompting project for vehicle component word expansion and input it into a preset Chinese large model, and obtain a vehicle component expansion word library corresponding to multiple vehicle components based on the output of the Chinese large model;

[0027] A mapping dictionary is determined according to an initial mapping dictionary and an extended vocabulary of vehicle components; in the mapping dictionary, vehicle components and component attributions are in a many-to-one mapping relationship.

[0028] In one embodiment, obtaining a demand text corresponding to a user demand includes:

[0029] The original demand text corresponding to the user demand is obtained, and at least one of the following cleaning is performed on the original demand text, and the cleaned text is used as the demand text:

[0030] Clear the target characters in the original appeal text;

[0031] Standardize the format of punctuation marks in the original appeal text;

[0032] Perform text correction on vehicle parts words in the original appeal text.

[0033] In one embodiment, the method further comprises:

[0034] When the output representation of the rule model identifies the attribution of the component, the attribution of the component is used as the initial attribution of the vehicle component; further, feature information corresponding to the appeal text is obtained, and the feature information is input into the classification model to obtain classification probability information;

[0035] When the probability corresponding to the initial vehicle part attribution in the classification probability information does not reach the preset probability condition, the initial vehicle part attribution is updated based on the classification probability information to obtain the vehicle part attribution corresponding to the user's demand.

[0036] In a second aspect, the present application also provides a vehicle component ownership determination device, comprising:

[0037] An acquisition module, used to acquire the demand text corresponding to the user's demand;

[0038] The first determination module is used to input the demand text into a preset rule model; the rule model is used to identify the attribution of the parts matching the demand text based on a preset mapping dictionary of vehicle parts and part attributions, and output the recognition result; when the output representation of the rule model identifies the attribution of the parts, the attribution of the parts is used as the attribution of the vehicle parts corresponding to the user demand;

[0039] The second determination module is used to obtain feature information corresponding to the demand text when the output representation of the rule model fails to identify the component ownership; input the feature information into a preset classification model, and determine the vehicle component ownership corresponding to the user's demand based on the classification probability information output by the classification model; the classification model is used to determine and output the probability that the feature information belongs to each component ownership among multiple preset component ownerships.

[0040] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the method for determining ownership of vehicle parts in the first aspect when executing the computer program.

[0041] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the method for determining ownership of vehicle parts in the first aspect are implemented.

[0042] In a fifth aspect, the present application further provides a computer program product, including a computer program, which, when executed by a processor, implements the steps in the method for determining ownership of vehicle parts in the first aspect.

[0043] The above-mentioned vehicle parts ownership determination method, device, computer equipment, computer-readable storage medium and computer program product use the rule model to identify the ownership of the parts that match the appeal text. If the part ownership is identified, the part ownership is used as the vehicle part ownership corresponding to the user's appeal; if the part ownership is not identified, the feature information corresponding to the appeal text is further obtained, and the feature information is input into the classification model, and the vehicle part ownership corresponding to the user's appeal is determined based on the classification probability information output by the classification model. This realizes the accurate determination of the ownership of the vehicle parts in the user's appeal. At the same time, because the vehicle parts ownership determination method combines the advantages of the fast speed of the rule model and the strong robustness of the classification model, the vehicle parts ownership determination method requires less time and has better efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A schematic diagram of a flow chart of a method for determining ownership of vehicle parts in one embodiment;

[0046] Figure 2 Another schematic diagram of a process for determining the ownership of vehicle parts in one embodiment;

[0047] Figure 3 It is another flowchart of a method for determining ownership of vehicle parts in one embodiment;

[0048] Figure 4 A schematic diagram of the principle of a method for determining attribution of vehicle parts in one embodiment;

[0049] Figure 5 is a structural block diagram of a vehicle component ownership determination device in one embodiment;

[0050] Figure 6FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] In an exemplary embodiment, Figure 1 As shown, a method for determining the ownership of vehicle parts is provided. In this embodiment, the execution subject of the method can be a device with data computing and processing functions, including but not limited to various personal computers, laptops, smart phones, tablet computers, servers, etc., and can also be a system of terminals and servers, etc. In short, as long as it has the information processing capabilities required to implement this method, its specific form is not limited. In this embodiment, the method includes the following steps S101 to S104. Among them:

[0053] Step S101: Obtain the demand text corresponding to the user's demand.

[0054] Among them, user demands, i.e., the demands regarding vehicles fed back by vehicle users, the original user demands can be in text form, or in the form of pictures, voice, video, etc., without limitation; the demand text, i.e., the text information reflecting the user demands obtained based on the user demands.

[0055] Step S102: input the appeal text into a preset rule model; the rule model is used to identify the component attribution matching the appeal text based on a preset mapping dictionary between vehicle components and component attributions, and output the recognition result.

[0056] Among them, vehicle parts can be understood as specific parts of the vehicle, such as windows, doors, tires, etc. Part ownership refers to the part to which the vehicle part should belong. For example, if the vehicle part is a door, then the part to which it should belong is the body, that is, the corresponding part belongs to the body. The mapping dictionary of vehicle parts and part ownership, that is, the pre-established mapping dictionary involving vehicle parts and part ownership, the mapping here is reflected in the correspondence between vehicle parts and part ownership. The rule model, that is, a pre-prepared artificial intelligence model for natural language processing, can be an existing artificial intelligence model, a model obtained by self-training, or a model obtained by targeted training based on an existing model. Based on the mapping dictionary, the rule model can identify the part ownership that matches the demand text and output the recognition result.

[0057] Step S103: When the output representation of the rule model identifies the attribution of the component, the attribution of the component is used as the attribution of the vehicle component corresponding to the user's request; when the output representation of the rule model does not identify the attribution of the component, characteristic information corresponding to the request text is obtained.

[0058] Among them, regarding the output form of the rule model: from the perspective of format, it can be text about the ownership of the identified components, or a table storing the ownership of the identified components, or a structured format such as JSON, XML, YAML, and plain text (unformatted), etc.; from the perspective of data structure, it can be a linear data structure, such as an array, a linked list, a stack, a queue, or other types of data structures, such as a tree data structure, a hash table; from the perspective of data type, it can be an integer, a floating point number, a string, etc. Exemplarily, the output of the rule model can be a list, which stores the ownership of the identified components. If the ownership of the components is not identified, the list is empty.

[0059] Among them, the characteristic information corresponding to the appeal text is information that can reflect, embody or contain one or two or more characteristics of the appeal text. Exemplarily, if the output representation of the rule model recognizes the attribution of the parts, that is, the recognition result contains the attribution of the parts, then the attribution of the parts can be used as the attribution of the vehicle parts corresponding to the user's appeal; conversely, if the output representation of the rule model does not recognize the attribution of the parts, that is, there is no attribution of the parts in the recognition result, then the characteristic information corresponding to the appeal text can be obtained to perform the subsequent steps of determining the attribution of the vehicle parts.

[0060] Step S104: Input the feature information into a preset classification model, and determine the vehicle component attribution corresponding to the user's request based on the classification probability information output by the classification model; the classification model is used to determine and output the probability that the feature information belongs to each component among multiple preset component attributions.

[0061] Among them, the classification model can be an existing artificial intelligence model, a model obtained by self-training, or a model obtained by targeted training based on an existing model. Regarding the output form of the classification model: from the perspective of format, it can be a text about the attribution of the identified components, or a table storing the attribution of the identified components, or a structured format such as JSON, XML, YAML, and plain text (unformatted), etc.; from the perspective of data structure, it can be a linear data structure, such as an array, a linked list, a stack, a queue, or other types of data structures, such as a tree data structure, a hash table; from the perspective of data type, it can be an integer, a floating point number, a string, etc. Correspondingly, the form of the classification probability information is the same as above. Exemplarily, the classification probability information can be a list, which can contain a series of floating point numbers, and these floating point numbers can represent the corresponding probability size, etc. Classification probability information, that is, the feature information output by the classification model belongs to the collection of probabilities of each component belonging to multiple preset component attributions. The classification model can determine the probability that the feature information belongs to each of the multiple preset component attributions and output it. Exemplarily, the feature information acquired in the previous step can be input into the preset classification model, and the classification model determines the probability that the feature information belongs to each of the multiple preset component attributions and outputs it, and then the vehicle component attribution corresponding to the user's request can be determined based on the classification probability information output by the classification model. For example, the higher the probability, the more likely the feature information is to belong to the component attribution corresponding to the probability, that is, the more likely the component attribution is the vehicle component attribution corresponding to the user's request, and then the component attribution corresponding to the highest probability can be taken as the vehicle component attribution corresponding to the user's request.

[0062] In this embodiment, the rule model is used to identify the attribution of the parts that match the appeal text. If the attribution of the parts is identified, the attribution of the parts is used as the attribution of the vehicle parts corresponding to the user's appeal; if the attribution of the parts is not identified, the attribution of the vehicle parts corresponding to the user's appeal is determined by obtaining the feature information corresponding to the appeal text and inputting the feature information into the classification model, based on the classification probability information output by the classification model. This realizes the determination of the attribution of the vehicle parts in the user's appeal. At the same time, because the vehicle parts attribution determination method combines the advantages of the fast speed of the rule model and the strong robustness of the classification model, the vehicle parts attribution determination method requires less time and has better efficiency and accuracy.

[0063] In one embodiment, if Figure 2As shown, a possible specific implementation method of "obtaining feature information corresponding to the appeal text" in step S103 is given, which specifically includes steps S201 to S203:

[0064] Step S201: extracting features from the appeal text to obtain the expression form features and semantic features of the appeal text.

[0065] Among them, the expression form characteristics of the appeal text, that is, the characteristics of the appeal text in terms of expression form, the expression form here is external to the text and does not involve the intrinsic meaning of the text. The expression form characteristics correspond to the semantic characteristics, which are the characteristics of the appeal text at two different levels: the expression form level and the semantic level. For example, the expression form characteristics and semantic characteristics of the appeal text can be obtained by extracting features from the appeal text to serve the subsequent steps of determining the feature information.

[0066] Step S202: Based on the expression form features, semantic features and their corresponding weights, the general features of the appeal text are obtained.

[0067] Exemplarily, after obtaining the expression features and semantic features through the aforementioned steps, weights can be assigned to the expression features and semantic features respectively, so as to obtain the general features of the appeal text according to the weights corresponding to the expression features and semantic features respectively, so as to serve the subsequent steps of determining the feature information.

[0068] Step S203: integrating the general features, expression form features and semantic features to determine the feature information corresponding to the appeal text.

[0069] Exemplarily, after obtaining the expression form features, semantic features and overview features through the aforementioned steps, the overview features, expression form features and semantic features may be fused to determine the feature information.

[0070] In this embodiment, the overview features are determined from the two levels of the expression form features and semantic features of the appeal text, and the feature information of the appeal text is determined from the three features of the expression form features, semantic features, and overview features. This enables the feature information to more accurately reflect the information carried by the appeal text, which improves the robustness of the method for determining the feature information and the entire method for determining the ownership of vehicle parts, and facilitates the subsequent determination of the ownership of vehicle parts based on the feature information, making the determination of the ownership of vehicle parts in user appeals more accurate.

[0071] In one of the embodiments, the expression form features of the appeal text include the pinyin features and glyph features of the appeal text; based on the expression form features, semantic features and their corresponding weights, the general features of the appeal text are obtained, including: taking the weighted average of the pinyin features, glyph features and semantic features as the general features of the appeal text; integrating the general features, expression form features and semantic features to determine the feature information corresponding to the appeal text, including: splicing the pinyin features, glyph features and semantic features to obtain the detail features corresponding to the appeal text; splicing the general features with the detail features to obtain the feature information of the appeal text.

[0072] Among them, the pinyin features of the appeal text, that is, the relevant features of the pinyin of the text involved in the appeal text, the pinyin features can be used to solve the text errors or text errors caused by different texts but the same or similar pronunciations. The glyph features of the appeal text, that is, the relevant features of the glyphs of the text involved in the appeal text, the glyph features can be used to solve the text errors or text errors caused by the similarity of the glyphs of the text. Exemplarily, the expression form features of the appeal text may include the pinyin features and glyph features of the appeal text, and then, the weighted average of the pinyin features, glyph features and semantic features can be calculated and used as the general features of the appeal text; the pinyin features, glyph features and semantic features can be connected in series to form a new feature about the appeal text-the detail features, and further, the obtained detail features can be connected in series with the general features obtained in the previous steps to form the feature information of the appeal text.

[0073] In this embodiment, the characteristic information of the appeal text is determined from two specific features, namely, pinyin features and glyph features. Since the characteristic information includes pinyin features, this helps to solve errors caused by different characters but the same or similar pronunciations; since the characteristic information includes glyph features, this helps to solve errors caused by similar glyphs of the characters. At the same time, by connecting the pinyin features, glyph features, and semantic features in series, and connecting the detail features with the overview features in series, the overview features are obtained by calculating the weighted average, and finally the characteristic information is obtained. Therefore, the characteristic information obtained in this embodiment can more accurately reflect the information carried by the appeal text and better reflect the user's real appeal, which is helpful for the subsequent analysis of the appeal text and other related steps, and further helps to make the ownership of the vehicle parts corresponding to the determined user appeal more accurate.

[0074] In one of the embodiments, feature extraction is performed on the appeal text to obtain the expression form features and semantic features of the appeal text, including: extracting the pinyin features of the appeal text based on the annotated pinyin data; extracting the glyph features of the appeal text based on the text glyphs; and extracting the semantic features of the appeal text based on a preset semantic analysis model.

[0075] Among them, the annotated pinyin data may include but is not limited to the basic composition of pinyin initials and finals, the correspondence between pinyin and Chinese characters, etc. The preset semantic analysis model is a pre-set artificial intelligence model with semantic analysis function, for example, the BERT (Bidirectional Encoder Representations from Transformers) model, which is a pre-trained language model based on the Transformer architecture. Exemplarily, the pinyin features of the appeal text can be extracted according to the annotated pinyin data. For example, a custom feature extraction algorithm can be used here to obtain the pinyin sequence corresponding to the appeal text and realize the effective conversion from Chinese characters to pinyin. Of course, the pinyin features can also be obtained with the help of an existing language model with feature extraction. Similarly, the glyph features of the appeal text can be extracted according to the text glyph. After obtaining the pinyin features and glyph features of the appeal text, the pinyin features and glyph features can be used as expression features, because the pinyin features and glyph features are both features of the appeal text from the perspective of expression. As for the semantic level of the appeal text, the preset semantic analysis model can be used to extract the semantic features of the appeal text.

[0076] In this embodiment, the pinyin features of the appeal text are extracted through the annotated pinyin data, and the glyph features of the appeal text are extracted through the text glyphs. The pinyin features and glyph features are used as expression form features, so that the expression form features can contain or reflect the characteristics of the appeal text in both pinyin and glyphs. The semantic features of the appeal text are obtained through the preset semantic analysis model, which helps to make the feature information of the appeal text determined later more accurate, better reflect the information carried by the appeal text, and better reflect the user's real appeal.

[0077] In one of the embodiments, the method further includes: constructing an initial mapping dictionary of vehicle parts and part attributions based on multiple vehicle parts and multiple part attributions; constructing a prompting project for vehicle part word expansion and inputting it into a preset Chinese large model, and obtaining a vehicle part expansion vocabulary corresponding to multiple vehicle parts based on the output of the Chinese large model; determining a mapping dictionary based on the initial mapping dictionary and the vehicle part expansion vocabulary; the vehicle parts and part attributions in the mapping dictionary are a many-to-one mapping relationship.

[0078] Among them, prompt engineering is an important branch in the field of natural language processing (NLP), which focuses on how to design and optimize prompts to improve the performance of pre-trained language models. A prompt is a language description that guides the model to generate a specific output. It can be a question, a sentence, or a paragraph. The preset Chinese big model, that is, a pre-set language big model with Chinese processing capabilities, can be an existing language big model, a language big model trained by itself, or a model obtained by targeted training based on an existing model.

[0079] Exemplarily, the vehicle component attribution determination method may include the construction of an initial mapping dictionary between vehicle components and component attribution. First, as many words as possible involving vehicle components are needed. Since the number of words that can be enumerated manually is limited, time-consuming, and has poor repeatability, the relevant technology of the language macro model can be used to mine relevant words. Specifically, an initial mapping dictionary between vehicle components and component attribution can be constructed first. The initial mapping dictionary can be relatively simple and have less content; then a prompt project for vehicle component word expansion is constructed and input into a preset Chinese macro model, that is, the Chinese macro model is allowed to expand vehicle component words and output according to the prompt project, thereby obtaining a vehicle component expansion vocabulary; finally, the mapping dictionary can be determined based on the initial mapping dictionary and the vehicle component expansion vocabulary, for example, the words in the vehicle component expansion vocabulary are supplemented to the initial mapping dictionary, thereby obtaining a mapping dictionary.

[0080] In this embodiment, considering the limited number of words that can be enumerated manually, time-consuming and poor in repeatability, a language model is used to expand and mine words related to vehicle parts, thereby providing the words required for mapping dictionary construction, which not only improves the efficiency of mapping dictionary construction and reduces related costs, but also enables the mapping dictionary to contain more words and cover a wider range, thereby helping to determine the attribution of vehicle parts based on the mapping dictionary to respond to more actual situations and determine the attribution of vehicle parts more accurately. At the same time, because the prompt engineering is used to guide the language model to expand the words related to vehicle parts, the number of words output by the model is larger, more accurate and comprehensive, which helps to build a better mapping dictionary, and thus better serve the determination of vehicle part attribution.

[0081] In one of the embodiments, obtaining a demand text corresponding to a user demand includes: obtaining an original demand text corresponding to the user demand, and performing at least one of the following cleaning operations on the original demand text, and using the cleaned text as the demand text: removing target characters in the original demand text; unifying the format of punctuation marks in the original demand text; and performing text correction on vehicle parts words in the original demand text.

[0082] Among them, the target characters may include but are not limited to HTML tags, spaces, line breaks, etc. Exemplarily, the original demand text corresponding to the user's demand can be obtained, and the original demand text can be cleaned to ensure the quality of the input text, and the cleaned text can be used as the demand text. The specific cleaning method can be any one or two or more of clearing the target characters in the original demand text, unifying the punctuation marks in the original demand text, and performing text error correction on the vehicle parts words in the original demand text. The specific implementation of the cleaning method can be achieved by a specific algorithm or by using a large language model, which is not limited here.

[0083] In this embodiment, by cleaning the original demand text corresponding to the user's demand, the quality of the input text is guaranteed, which serves the subsequent analysis of the demand text and other related steps; at the same time, by correcting the vehicle parts words in the original demand text, the demand text can more accurately reflect the true intention of the user's demand, and avoid the subsequent analysis of the demand text and other related steps affected by the poor quality of the demand text, which helps to ensure the accurate determination of the ownership of vehicle parts.

[0084] In one embodiment, if Figure 3 As shown, the method may further include steps S301 to S303:

[0085] Step S301: When the output representation of the rule model identifies the component attribution, the component attribution is used as the initial vehicle component attribution.

[0086] For example, if the rule model can identify the part ownership corresponding to the claim text, then in order to ensure that the determined part ownership has a higher accuracy, the identified part ownership can be used as the initial vehicle part ownership, and then the accuracy of the initial vehicle part ownership can be verified through subsequent steps.

[0087] Step S302: further acquiring feature information corresponding to the appeal text, and inputting the feature information into a classification model to obtain classification probability information.

[0088] Exemplarily, to verify the accuracy of the initial vehicle parts attribution, the feature information corresponding to the appeal text can be obtained, and the feature information can be input into the classification model to obtain classification probability information. The specific implementation of this step can be found in the relevant content of the aforementioned embodiment, which will not be repeated here.

[0089] Step S303: When the probability corresponding to the initial vehicle component attribution in the classification probability information does not reach the preset probability condition, the initial vehicle component attribution is updated based on the classification probability information to obtain the vehicle component attribution corresponding to the user's request.

[0090] For example, after obtaining the classification probability information, the accuracy of the initial vehicle component attribution can be verified according to the classification probability information. If the verification result is that the preset probability condition is not met, the initial vehicle component attribution is updated as the vehicle component attribution corresponding to the user's request based on the classification probability information.

[0091] Exemplarily, for the above verification, a probability condition can be preset. If the probability corresponding to the initial vehicle part attribution does not meet the probability condition, it means that the accuracy of the initial vehicle part attribution is not good. In this regard, the initial vehicle part attribution can be updated as the vehicle part attribution corresponding to the user's request based on the classification probability information. For example, the initial vehicle part attribution is "body", and the probability corresponding to "body" in the classification probability information is 30%, and the probability required by the preset probability condition is 60%. At this time, the probability corresponding to the initial vehicle part attribution does not meet the probability condition, that is, the verification result is that the preset probability condition is not met; at the same time, in the classification probability information, the highest probability is 89%, and the part corresponding to 89% is "chassis". Then, the initial vehicle part attribution can be updated as the vehicle part attribution corresponding to the user's request, that is, "chassis" can be used as the vehicle part attribution corresponding to the user's request. Here is only an example of a possible way to update the initial vehicle parts attribution as the vehicle parts attribution corresponding to the user's request based on the classification probability information. Of course, there can be other ways, for example, weighting each probability in the classification probability information, removing the highest value and the lowest value, and then updating the initial vehicle parts attribution as the vehicle parts attribution corresponding to the user's request based on the part attribution corresponding to the highest probability after processing.

[0092] In this embodiment, by obtaining the feature information corresponding to the appeal text and inputting the feature information into the classification model, the classification probability information is obtained, and the accuracy of the initial vehicle parts attribution is verified by using the classification probability information. If the verification result is that the preset probability condition is not met, the initial vehicle parts attribution can be updated as the vehicle parts attribution corresponding to the user appeal based on the classification probability information. This makes the determined vehicle parts attribution more accurate. At the same time, the initial vehicle parts attribution is obtained through the rule model, which ensures the speed of determining the vehicle parts attribution; through the above verification and update of the initial vehicle parts attribution, the accuracy of determining the vehicle parts attribution is guaranteed.

[0093] In an exemplary embodiment, a method for determining the ownership of vehicle parts is provided, and the overall framework and detailed process of the method are as follows:

[0094] 1. The overall framework of this method

[0095] like Figure 4As shown, the overall framework of the vehicle parts attribution determination method in this embodiment is given. The execution subject of the method can be a system for determining the attribution of vehicle parts. When the system obtains the text of the user's request, the text cleaning module in the system is responsible for cleaning the text data to ensure the quality of the text; then the cleaned text data is submitted to the rule model, and the rule model identifies the attribution of the parts of the text. If the rule model can effectively identify the attribution of the parts, then the attribution of the parts is directly output; if the rule model cannot effectively identify the attribution of the parts, then enter the sample construction module, build a sample that conforms to the text classification model based on the request text, and then feed the sample to the classification model, and the model gives the attribution of the parts of the text. Compared with the traditional method, this method combines the advantages of the two models by first using the rule model processing and then calling the classification model strategy, and maximizes the efficiency of part attribution recognition while ensuring accuracy.

[0096] 2. Detailed process of this method

[0097] The detailed process of this method can include the following four steps:

[0098] Step 1: Text cleaning. Text cleaning methods include unifying punctuation marks, removing target characters (HTML tags, spaces, line breaks, etc.). In line with the goals of this project, we introduce Chinese word error correction technology based on the above text cleaning methods to improve text quality. Here is a possible specific method for text cleaning:

[0099] (1) Clearing of blank characters, including spaces, tab spaces, and other character spaces. For example, the text before processing is: "The user called to say that the vehicle was charging and prompted that the power system was limited. Restarting the car computer and powering on and off did not work. He asked what to do." The text after processing is: "The user called to say that the vehicle was charging and prompted that the power system was limited. Restarting the car computer and powering on and off did not work. He asked what to do."

[0100] (2) Punctuation is unified, and is unified into English punctuation. For example, the text before processing is: "User calls: saying that when parking on the night of February 29, the front of the car hit a pillar, no one was injured, and the video dashcam did not record any relevant video from the time of the incident between 21:16 and 21:30, and AEB was not triggered at the time. The user said that he needed to retrieve the background data and needed to know the reason." The text after processing is: "User calls: saying that when parking on the night of February 29, the front of the car hit a pillar, no one was injured, and the video dashcam did not record any relevant video from the time of the incident between 21:16 and 21:30, and AEB was not triggered at the time. The user said that he needed to retrieve the background data and needed to know the reason."

[0101] (3) Text error correction: Keywords related to the attribution of vehicle parts are very important. If typos are found, they need to be corrected to improve the model recognition ability. For example, the text before processing is: "The user came in to report the NCA upgrade problem, and reported that the main driver's window made noise when the car ran to about 120 mph"; the text after processing is: "The user came in to report the NCA upgrade problem, and reported that the car ran to about 120 mph and the main driver's window made noise."

[0102] Step 2: Use the rule model to determine the ownership of vehicle parts. Before this, the corresponding mapping dictionary can be pre-built. The following is a possible mapping dictionary construction method, which corresponds to "based on multiple vehicle parts and multiple part ownerships, build an initial mapping dictionary for vehicle parts and part ownership; build a prompt project for vehicle part word expansion and input it into the preset Chinese large model, and based on the output of the Chinese large model, obtain a vehicle part expansion vocabulary corresponding to multiple vehicle parts; determine the mapping dictionary based on the initial mapping dictionary and the vehicle part expansion vocabulary; the vehicle parts and part ownership in the mapping dictionary are a many-to-one mapping relationship" and other related contents, as follows:

[0103] First, we manually collected words that describe automobile parts and built a preliminary vocabulary. The vocabulary template example is as follows:

[0104] {

[0105] 'Body surface covering': 'Body',

[0106] 'Window Glass': 'Body',

[0107] 'Skylight': 'Body',

[0108] 'Back door': 'Body',

[0109] 'Door': 'Body',

[0110] 'Suspension': 'Chassis',

[0111] 'Steering': 'Chassis',

[0112] 'Brake': 'Chassis',

[0113] 'Hub': 'Chassis',

[0114] 'Motor': 'Power',

[0115] 'Car Charging': 'Power',

[0116] 'Driving mode': 'Power'

[0117] }

[0118] For each row, the first word is a word describing a car function or component, and the second word is the car module to which the vehicle component should belong, that is, the component attribution. The relationship between the words in the first column and the words in the second column is many-to-one. The requirements for building a vocabulary are: the more words, the better, the more complete the word expressions, the better, and the wider the word coverage, the better.

[0119] Then, based on the big model technology, the synonyms and antonyms of automobile parts are mined to expand the vocabulary. Due to the limited number of words that can be enumerated manually, the time-consuming nature and the poor repeatability, this method uses the big model technology to mine related words. (1) Design a prompt. Prompt example: "You are a product expert in the automotive field. Based on the following words: word 1, word 2, word 3..., from the perspective of a car user, try to give the synonyms, antonyms or similar words of each word." Word 1, word 2, and word 3 in the prompt are the words in the first column of the vocabulary in step 1, such as: sunroof, car door, wheel hub. (2) Input the designed prompt into the big model with Chinese capabilities and let the big model expand the words. (3) Sort out the new words output by the large model. The sorting results are as follows: {"roof glass":"skylight","skylight glass":"skylight","roof window":"skylight","skylighting skylight":"skylight","roof lighting panel":"skylight","skylighting window":"skylight","roof transparent window":"skylight","door frame":"door","door lock":"door"} Among the new words, roof glass is another description of skylight.

[0120] Finally, the words obtained in the previous steps are integrated to build a massive vocabulary that can describe automobile parts, and a mapping dictionary is built from words to vehicle parts and part attributes. An example of the newly constructed dictionary is as follows:

[0121] {

[0122] 'Body surface covering': 'Body',

[0123] 'Window Glass': 'Body',

[0124] 'Skylight': 'Body',

[0125] 'Roof glass': 'Body',

[0126] 'Roof Window': 'Body',

[0127] 'Roof transparent window': 'Body',

[0128] 'Door frame': 'Body',

[0129] 'Back door': 'Body',

[0130] 'Door': 'Body',

[0131] 'Suspension': 'Chassis',

[0132] 'Steering': 'Chassis',

[0133] 'Brake': 'Chassis',

[0134] 'Hub': 'Chassis',

[0135] 'Motor': 'Power',

[0136] 'Car Charging': 'Power',

[0137] 'Driving mode': 'Power'

[0138] }

[0139] Among them, "'roof glass':'body',

[0140] 'Roof Window': 'Body',

[0141] 'Roof transparent window': 'Body',

[0142] 'Door frame':'car body'," is the expanded part.

[0143] After obtaining the above mapping dictionary, the rule model can be used to determine the ownership of vehicle parts. This step corresponds to "inputting the demand text into the preset rule model; the rule model is used to identify the ownership of parts matching the demand text based on the preset mapping dictionary of vehicle parts and parts ownership, and output the recognition result" and other related content. The details are as follows:

[0144] The cleaned text is used as the appeal text and input into the rule model to determine the attribution of vehicle parts. The rule model can extract words related to vehicle parts from the appeal text based on text matching technology. These words can be words in the preset mapping dictionary of vehicle parts and part attribution. Specifically, the rule model can traverse each vehicle part word in the vocabulary (such as sunroof, roof glass), and for each word, use the bidirectional maximum matching algorithm to determine whether the appeal text contains the word. If the word is contained, the word is stored in the data queue queue, that is, the recognition result. The example is as follows: User appeal: "The user reported that the vehicle ran to about 120 yards and the main driver's window had noise", the corresponding matching result: queue=["window"]. After obtaining the matching result, the rule model can traverse each word in the matching result, map each word to the part attribution in the mapping dictionary according to the aforementioned mapping dictionary, and then output the corresponding part attribution belongList. For example, user demand: "The user reported that the car was running at about 120 mph and there was noise from the main driver's window"; matching result: queue=["window"]; mapping result: belongList=["body"].

[0145] The principle of this step is to find out the vehicle parts involved in the appeal text, such as windows, tires, etc., and then find the corresponding parts from the mapping dictionary according to the vehicle parts involved. For example, in the mapping dictionary, the parts corresponding to vehicle parts such as "doors", "window glass", and "sunroof" are all "body", that is, the parts with mapping relationships with vehicle parts such as "doors", "window glass", and "sunroof" are "body". Therefore, when the vehicle part involved in the appeal text is "door", the corresponding part is "body".

[0146] Step 3: Determine whether the rule model has determined the ownership of vehicle parts. This corresponds to "when the output representation of the rule model identifies the ownership of parts, the ownership of parts is used as the ownership of vehicle parts corresponding to the user's request; when the output representation of the rule model does not identify the ownership of parts, the feature information corresponding to the request text is obtained" and other related content. The details are as follows:

[0147] If the rule model outputs clear part attribution, that is, belongList contains at least one part attribution word; then the corresponding vehicle part attribution can be output; if no clear part attribution can be given, that is, there is no part attribution word in belongList, then the classification model can be used to determine the vehicle part attribution based on semantic judgment.

[0148] Step 4: Use the classification model to determine the attribution of vehicle parts. This step can include two steps: sample construction and analysis based on the constructed sample using the classification model, as follows:

[0149] Regarding the sample construction step, it corresponds to "obtaining the feature information corresponding to the appeal text" and other related content. The sample here refers to the feature information. This step specifically includes:

[0150] Pinyin annotation, based on the annotated pinyin, extracts distributed features, namely pinyin features, to assist the model in solving homophone problems. An example is as follows: User demand: "The user came in to report the NCA upgrade problem, and reported that the main driver's lathe made noise when the car ran to about 120 mph"; The real demand that the user wants to express: "The user came in to report the NCA upgrade problem, and reported that the car ran to about 120 mph, and the main driver's window made noise"; Distributed feature representation: f1=[0.3530673,0.22740743,0.30655342,0.77849578,0.25106031].

[0151] Extract the glyph features of the text to solve the problem of similar characters and assist the model in understanding the semantics of the text. The following is an example: User demand: "The user called to say that after passing the mountain road, the wheel hub had cracks"; The real demand that the user wants to express: "The user called to say that after passing the mountain road, the wheel hub had cracks"; Distributed feature representation: f2=[0.86786499,0.79189034,0.33742122,0.37105652,0.90973686].

[0152] Based on the BERT model, the semantic features of the text are extracted to assist the model in understanding the meaning of words in the text.

[0153] The phonetic features, glyph features, and semantic features in the above steps are weighted averaged to obtain the general features; the features are connected in series to form the detailed features; finally, the detailed features and the general features are connected in series to form a full and detailed feature space including the pronunciation, word form, and semantics, i.e., feature information. An example is as follows:

[0154] User’s claim: “A user called and said that when parking his car on the night of February 29, the front of the car hit a pillar, and no one was injured. There was no relevant record on the dashcam during the period, no relevant video, and AEB was not triggered at the time.”

[0155] Pinyin features, f1=[0.72550418, 0.99530686, 0.54699301, 0.56576056, 0.09869959]; glyph features, f2=[0.91131963, 0.54507588, 0.28106904, 0.76675673, 0.68355904]; semantic features, f3=[0.41762336, 0.72809674, 0.1629559, 0.53736758, 0.76744826].

[0156] The calculation formula for the overview feature is: (f1+f2+f3) / 3, and the calculation result is: feature_overiew=[0.68481573,0.75615983,0.33033932,0.62329496,0.51656896].

[0157] Features are concatenated to form detail features:

[0158] feature_detail=[0.72550418,0.99530686,0.54699301,0.56576056,0.09869959,0.91131963,0.545075 88,0.28106904,0.76675673,0.68355904,0.41762336,0.72809674,0.1629559,0.53736758,0.76744826].

[0159] The general features and detailed features are connected in series to construct the final feature space, i.e., feature information:

[0160] feature=[0.68481573,0.75615983,0.33033932,0.62329496,0.51656896,0.72550418,0.99530686,0.54699301,0.56576056,0.0986 9959,0.91131963,0.54507588,0.28106904,0.76675673,0.68355904,0.41762336,0.72809674,0.1629559,0.53736758,0.76744826].

[0161] Regarding the step of using the classification model to analyze the constructed samples, it corresponds to "inputting the feature information into the preset classification model, determining the vehicle component attribution corresponding to the user's demand based on the classification probability information output by the classification model; the classification model is used to determine the probability of the feature information belonging to each component in multiple preset component attributions and output it" and other related contents, specifically including:

[0162] The feature information extracted in the previous steps is fed to the trained classification model to predict the component belonging to the appeal text, and the category with the largest predicted probability is selected as the output. For example, suppose there are only 5 categories for the component belonging ["battery", "trim", "cockpit", "body", "chassis"]; user appeal: "The user called in to report the NCA upgrade problem, and reported that the main driver's window made noise when the car ran to about 120 yards"; the model output results: [0.02, 0.12, 0.25, 0.6, 0.01]. The value at each position represents the probability of the category represented by the position. From the output results, it can be seen that the probability at position 4 is the largest, which is 0.6. Then, it can be judged that the component corresponding to the user's appeal belongs to "body".

[0163] The vehicle parts attribution determination method provided in this embodiment combines the characteristics of fast rule model and strong robustness of classification model, and improves the speed and accuracy of part attribution identification. In the text classification model, a new feature space, namely feature information, is designed to improve the robustness of the algorithm. In the process of designing the rule model, a large number of words that can describe automobile parts are established by using large model technology, which reduces the cost of word library construction and improves the comprehensiveness of the rules and the ability to deal with complex texts.

[0164] In this embodiment, since the vehicle parts ownership determination method combines the advantages of fast rule model speed and strong classification model robustness, the vehicle parts ownership determination method takes less time and has better efficiency and accuracy. The feature information of the appeal text is determined from two specific features, namely, pinyin feature and glyph feature. Since the feature information has pinyin feature, it helps to solve errors caused by different words but the same or similar pronunciation; since the feature information has glyph feature, it helps to solve errors caused by similar glyphs. At the same time, feature information is obtained by connecting pinyin feature, glyph feature and semantic feature in series, and connecting detail feature and general feature in series. Therefore, the feature information can more accurately reflect the information carried by the appeal text and the real appeal of the user, which is helpful for the subsequent analysis of the appeal text and other related steps, and thus helps to make the ownership of the vehicle parts corresponding to the determined user appeal more accurate. Considering the limited number of words that can be enumerated manually, time-consuming and poor in repeatability, the language model is used to expand and mine the words related to vehicle parts, so as to provide the words needed for mapping dictionary construction. This not only improves the efficiency of mapping dictionary construction and reduces related costs, but also enables the mapping dictionary to contain more words and cover a wider range, which helps to determine the attribution of vehicle parts according to the mapping dictionary. The actual situation can be more and the determined attribution of vehicle parts is more accurate. At the same time, because the prompt engineering is used to guide the language model to expand the words related to vehicle parts, the number of words output by the model is larger, more accurate and comprehensive, which helps to build a better mapping dictionary, and thus better serve the determination of vehicle parts attribution. Through text cleaning, the text quality is guaranteed and the subsequent steps such as analysis of the appeal text are carried out; at the same time, by correcting the words of vehicle parts in the text, the appeal text can more accurately reflect the true intention of the user's appeal, and avoid the subsequent steps such as analysis of the appeal text affected by the poor quality of the text, which helps to ensure the accurate determination of the attribution of vehicle parts.

[0165] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0166] Based on the same inventive concept, the embodiment of the present application also provides a vehicle component attribution determination device for implementing the vehicle component attribution determination method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more vehicle component attribution determination device embodiments provided below can refer to the limitations of the vehicle component attribution determination method above, and will not be repeated here.

[0167] In an exemplary embodiment, Figure 5 As shown, a vehicle component attribution determination device 500 is provided, comprising:

[0168] An acquisition module 501 is used to acquire a demand text corresponding to a user demand;

[0169] The first determination module 502 is used to input the demand text into a preset rule model; the rule model is used to identify the attribution of the component matching the demand text based on a preset mapping dictionary of vehicle components and component attributions, and output the identification result; when the output representation of the rule model identifies the attribution of the component, the attribution of the component is used as the attribution of the vehicle component corresponding to the user demand;

[0170] The second determination module 503 is used to obtain feature information corresponding to the demand text when the output representation of the rule model fails to identify the component ownership; input the feature information into a preset classification model, and determine the vehicle component ownership corresponding to the user demand based on the classification probability information output by the classification model; the classification model is used to determine and output the probability that the feature information belongs to each component ownership among multiple preset component ownerships.

[0171] In one of the embodiments, the second determination module 503 is also used to extract features from the appeal text to obtain the expression features and semantic features of the appeal text; obtain the general features of the appeal text based on the expression features, semantic features and their corresponding weights; and determine the feature information corresponding to the appeal text by integrating the general features, expression features and semantic features.

[0172] In one of the embodiments, the expression form features of the appeal text include the phonetic features and glyph features of the appeal text; the second determination module 503 is further used to take the weighted average of the corresponding phonetic features, glyph features and semantic features as the general features of the appeal text; concatenate the phonetic features, glyph features and semantic features to obtain the detailed features corresponding to the appeal text; and concatenate the general features with the detailed features to obtain the feature information of the appeal text.

[0173] In one embodiment, the second determination module 503 is further used to extract the pinyin features of the appeal text based on the annotated pinyin data; extract the glyph features of the appeal text based on the text glyphs; and extract the semantic features of the appeal text based on a preset semantic analysis model.

[0174] In one embodiment, the first determination module 502 is further used to construct an initial mapping dictionary of vehicle parts and part attributions based on multiple vehicle parts and multiple part attributions; construct a prompt project for vehicle part word expansion and input it into a preset Chinese large model, and based on the output of the Chinese large model, obtain a vehicle part expansion vocabulary corresponding to multiple vehicle parts; determine a mapping dictionary based on the initial mapping dictionary and the vehicle part expansion vocabulary; the vehicle parts and part attributions in the mapping dictionary are a many-to-one mapping relationship.

[0175] In one of the embodiments, the acquisition module 501 is also used to obtain the original demand text corresponding to the user's demand, and perform at least one of the following cleaning on the original demand text, and use the cleaned text as the demand text: clear the format of the target characters in the original demand text; unify the punctuation marks in the original demand text; perform text correction on the vehicle parts words in the original demand text.

[0176] In one embodiment, the first determination module 502 is further configured to use the component attribution as the initial vehicle component attribution when the output representation of the rule model identifies the component attribution;

[0177] The second determination module 503 is further used to further obtain feature information corresponding to the appeal text, and input the feature information into the classification model to obtain classification probability information;

[0178] The first determination module 502 is further configured to update the initial vehicle component attribution based on the classification probability information to obtain the vehicle component attribution corresponding to the user's request when the probability corresponding to the initial vehicle component attribution in the classification probability information does not meet a preset probability condition.

[0179] Each module in the above-mentioned vehicle component ownership determination device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0180] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data required for determining the ownership of vehicle parts, such as a mapping dictionary. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for determining the ownership of vehicle parts is implemented.

[0181] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0182] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the vehicle component ownership determination method in the aforementioned embodiment are implemented.

[0183] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the vehicle component ownership determination method in the aforementioned embodiment are implemented.

[0184] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the vehicle component ownership determination method in the above-mentioned embodiment when executed by a processor.

[0185] It should be noted that the user information (including but not limited to user device information, user personal information, information related to user demands, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0186] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0187] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0188] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for determining the ownership of vehicle parts, characterized in that: include: Get the demand text corresponding to the user's demand; Inputting the appeal text into a preset rule model; the rule model is used to identify the attribution of the components matching the appeal text based on a preset mapping dictionary of vehicle components and component attributions, and output the recognition result; In the case where the output representation of the rule model identifies the attribution of a component, using the attribution of the component as the attribution of the vehicle component corresponding to the user demand; When the output representation of the rule model fails to identify the attribution of the component, obtaining feature information corresponding to the appeal text; The feature information is input into a preset classification model, and based on the classification probability information output by the classification model, the vehicle component attribution corresponding to the user demand is determined; the classification model is used to determine and output the probability that the feature information belongs to each component among multiple preset component attributions.

2. The method according to claim 1, characterized in that The obtaining of feature information corresponding to the appeal text includes: Extracting features of the appeal text to obtain expression form features and semantic features of the appeal text; Based on the expression form features, the semantic features and their corresponding weights, obtaining the general features of the appeal text; The overview features, the expression form features and the semantic features are integrated to determine feature information corresponding to the appeal text.

3. The method according to claim 2, characterized in that The expression form characteristics of the appeal text include the pinyin characteristics and the glyph characteristics of the appeal text; The obtaining of the general features of the appeal text based on the expression form features, the semantic features and their corresponding weights includes: The pinyin feature, the glyph feature and the semantic feature are weighted averaged according to their respective corresponding weights to obtain the general features of the appeal text; The fusing of the overview feature, the expression form feature and the semantic feature to determine the feature information corresponding to the appeal text includes: The pinyin feature, the glyph feature and the semantic feature are combined to obtain the detail feature corresponding to the appeal text; The general features and the detailed features are combined to obtain the feature information of the appeal text.

4. The method according to claim 3, characterized in that The feature extraction of the appeal text to obtain the expression form features and semantic features of the appeal text includes: Extracting the pinyin features of the appeal text based on the annotated pinyin data; Based on the text font, extracting font features of the appeal text; Based on a preset semantic analysis model, the semantic features of the appeal text are extracted.

5. The method according to claim 1, characterized in that The method further comprises: Based on multiple vehicle parts and multiple part attributions, construct an initial mapping dictionary between vehicle parts and part attributions; Constructing a prompting project for vehicle component word expansion and inputting it into a preset Chinese macro model, and obtaining a vehicle component expansion vocabulary corresponding to the plurality of vehicle components based on the output of the Chinese macro model; The mapping dictionary is determined according to the initial mapping dictionary and the vehicle component expansion vocabulary; the vehicle components and component attributions in the mapping dictionary are in a many-to-one mapping relationship.

6. The method according to any one of claims 1 to 5, characterized in that: The obtaining of the demand text corresponding to the user demand includes: The original demand text corresponding to the user demand is obtained, and at least one of the following cleaning is performed on the original demand text, and the cleaned text is used as the demand text: Clearing target characters from the original appeal text; Unify the format of punctuation marks in the original appeal text; Perform text correction on vehicle parts words in the original appeal text.

7. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: In the case where the output representation of the rule model identifies the attribution of the component, the attribution of the component is used as the initial attribution of the vehicle component; further acquiring feature information corresponding to the appeal text, and inputting the feature information into the classification model to obtain classification probability information; When the probability corresponding to the initial vehicle component attribution in the classification probability information does not meet the preset probability condition, the initial vehicle component attribution is updated based on the classification probability information to obtain the vehicle component attribution corresponding to the user demand.

8. A vehicle component ownership determination device, characterized in that: The device comprises: An acquisition module, used to acquire the demand text corresponding to the user's demand; A first determination module is used to input the demand text into a preset rule model; the rule model is used to identify the component attribution matching the demand text based on a preset mapping dictionary between vehicle components and component attributions, and output a recognition result; when the output representation of the rule model identifies the component attribution, the component attribution is used as the vehicle component attribution corresponding to the user demand; The second determination module is used to obtain the feature information corresponding to the demand text when the output representation of the rule model fails to identify the component ownership; input the feature information into a preset classification model, and determine the vehicle component ownership corresponding to the user demand based on the classification probability information output by the classification model; the classification model is used to determine and output the probability that the feature information belongs to each component ownership among multiple preset component ownerships.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images