Item category recognition method, device, electronic device and storage medium

Abstract

An embodiment of the present invention discloses an article category recognition method, device, electronic device, and storage medium. The method includes: obtaining description information of an article to be recognized, and obtaining the original category to which the article to be recognized belongs in the original category system; querying a preset mapping relationship based on the original category to obtain a first standard category to which the article to be recognized belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than that in the original category system; inputting the description information into a category recognition model to obtain a second standard category to which the article to be recognized belongs in the standard category system; determining a target standard category to which the article to be recognized belongs in the standard category system based on the first standard category and / or the second standard category. The embodiment of the present invention combines category system mapping to recognize categories and model recognition of categories, realizes automatic recognition of article categories, has high recognition efficiency, and reduces labor costs.

Description

Technical Field

[0001] The present invention relates to item classification technology, and in particular, to an item category recognition method, device, electronic device, and storage medium. Background Art

[0002] Items, especially industrial items, due to the characteristics of a large variety of categories, high specialization, and small category subdivision granularity, etc., lead to the fact that the category recognition methods in other fields (such as industry category recognition) cannot achieve applicable effects in the item field. In the process of implementing the present invention, the inventors found that the current item category recognition cannot be carried out by machine mounting and all rely on category experts of different categories to manually identify categories. Manual item category recognition has problems such as low efficiency and high labor costs. Summary of the Invention

[0003] Embodiments of the present invention provide an item category recognition method, device, electronic device, and storage medium, which can realize automatic recognition of item categories, have high accuracy, improve category recognition efficiency, and reduce labor costs.

[0004] In a first aspect, an embodiment of the present invention provides an item category recognition method, including:

[0005] Obtain the description information of the item to be recognized, and obtain the original category to which the item to be recognized belongs in the original category system;

[0006] Query a preset mapping relationship based on the original category to obtain a first standard category to which the item to be recognized belongs in the standard category system. The preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than the category classification granularity in the original category system;

[0007] Input the description information into a category recognition model for category recognition to obtain a second standard category to which the item to be recognized belongs in the standard category system;

[0008] Determine a target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and / or the second standard category.

[0009] In a second aspect, an embodiment of the present invention provides an item category recognition device, including:

[0010] An obtaining module, configured to obtain the description information of the item to be recognized, and obtain the original category to which the item to be recognized belongs in the original category system;

[0011] A query module, configured to query a preset mapping relationship based on the original category to obtain a first standard category to which the item to be recognized belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than that in the original category system;

[0012] An identification module, configured to input the description information into a category identification model for category identification to obtain a second standard category to which the item to be recognized belongs in the standard category system;

[0013] A determination module, configured to determine a target standard category to which the item to be recognized belongs in the standard category system based on the first standard category or the second standard category.

[0014] In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the item category identification method as described in any one of the embodiments of the present invention is implemented.

[0015] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the item category identification method as described in any one of the embodiments of the present invention is implemented.

[0016] In the embodiments of the present invention, the description information of the item to be recognized can be obtained, and the original category to which the item to be recognized belongs in the original category system can be obtained; a preset mapping relationship is queried based on the original category to obtain a first standard category to which the item to be recognized belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than that in the original category system; the description information is input into a category identification model for category identification to obtain a second standard category to which the item to be recognized belongs in the standard category system; a target standard category to which the item to be recognized belongs in the standard category system is determined based on the first standard category and / or the second standard category. That is, in the embodiments of the present invention, the method of mapping the category system to identify categories and the method of using a model to identify categories are combined to implement the category identification of items. The entire category identification method can be implemented by machine mounting, automatically, with high identification efficiency and reduced labor costs; in addition, by combining the method of mapping the category system to identify categories and the method of using a model to identify categories, the accuracy of category identification is improved, and more fine-grained category identification can be achieved. Description of the Drawings

[0017] Figure 1 is a flowchart of an item category identification method provided by an embodiment of the present invention;

[0018] Figure 2 It is a schematic structural diagram of the category recognition model provided by the embodiments of the present invention;

[0019] Figure 3 It is a schematic flowchart of the method for identifying categories by using the category system mapping provided by the embodiments of the present invention;

[0020] Figure 4 It is a schematic flowchart of the method for identifying categories by using the model provided by the embodiments of the present invention;

[0021] Figure 5 It is another schematic flowchart of the article category recognition method provided by the embodiments of the present invention;

[0022] Figure 6 It is an example diagram of the article category recognition method provided by the embodiments of the present invention;

[0023] Figure 7 It is a schematic structural diagram of the article category recognition device provided by the embodiments of the present invention;

[0024] Figure 8 It is a schematic structural diagram of the electronic device provided by the embodiments of the present invention. Detailed implementation manners

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. In addition, it should be noted that, for the sake of convenience of description, only parts related to the present invention rather than all structures are shown in the accompanying drawings.

[0026] Figure 1 It is a schematic flowchart of the article category recognition method provided by the embodiments of the present invention. This method can be executed by the article category recognition device provided by the embodiments of the present invention, and the device can be implemented in a software and / or hardware manner. In a specific embodiment, the device can be integrated in an electronic device, such as a computer, a server, etc. The following embodiments will be described by taking the device integrated in the electronic device as an example. Refer to Figure 1 , and the method can specifically include the following steps:

[0027] Step 101, obtain the description information of the item to be recognized, and obtain the original category to which the item to be recognized belongs in the original category system.

[0028] Exemplarily, the item to be recognized can be any item that needs to be classified, and this item can be an industrial product (such as valves, pipe fittings, instruments, tools, etc.), or a non-industrial product (such as mobile phones, refrigerators, washing machines, etc.), and no specific limitation is made here. The description information of the item to be recognized, such as the name, title, etc. of the item to be recognized, can be description information in text form. For example, when it is necessary to classify an item listed on a certain website, the name, title, etc. of the item can be captured from the web page of the website to obtain the description information of the item.

[0029] In practical applications, with the continuous change of classification requirements, there may be different category systems, such as the original category system and the standard category system. The original category system can be the category system used previously, and the standard category system can be the category system currently in use or about to be put into use. Specifically, in the embodiments of the present invention, the category classification granularity under the standard category system can be finer than that under the original category system. For example, the item classification under the original category system is divided into three levels: primary category, secondary category, and terminal category; the item classification under the standard category system is divided into four levels: primary category, secondary category, sub-terminal category, and terminal category; compared with the original category system, the standard category system classifies items more finely. It should be noted that for categories with the same level under the original category system and the standard category system, the specific classification content may be different; for example, the secondary category of the original category system may be manual tools, and the secondary category of the standard category system may be cutting tools. In the embodiments of the present invention, the original category to which the item to be recognized belongs under the original category system can be pre-recognized and stored, and can be directly obtained from the storage when needed.

[0030] Step 102, query the preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs under the standard category system. The preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity under the standard category system is finer than that under the original category system.

[0031] Exemplarily, the preset mapping relationship can be established by manually or machine-analyzing the original category system and the standard category system. Using this preset mapping relationship, any original category under the original category system can be queried for the corresponding standard category under the standard category system. Since the category classification granularity under the standard category system is finer than that under the original category system, therefore, any original category under the original category system queried using this preset mapping relationship may have multiple corresponding standard categories under the standard category system. These multiple standard categories can form a standard category set, and the standard categories in the standard category set can be represented by the terminal categories under the standard category system, or can be represented by the complete level categories under the standard category system.

[0032] Specifically, in the embodiments of the present invention, there are also multiple standard categories to which the to-be-recognized items recalled based on the original category to which the to-be-recognized items belong using the preset mapping relationship belong in the standard category system. The multiple standard categories form a standard category set, and a standard category can be selected from the standard category set as the first standard category according to rules.

[0033] Step 103: Input the description information into the category recognition model for category recognition to obtain the second standard category to which the to-be-recognized item belongs in the standard category system.

[0034] Specifically, the category recognition model is used to recognize the standard category to which an item belongs in the standard category system. The category recognition model can be trained using a preset data set. The sample data in the preset data set can be the description information of a large number of items, and the sample labels can be the standard categories to which the corresponding items belong in the standard category system. The sample labels can be generated by manually or machine-labeling the sample data.

[0035] In specific applications, the preset data set can be divided into a training set and a validation set according to the data volume ratio. For example, the data volume ratio of the training set to the validation set can be 8:2, 9:1, etc.

[0036] Exemplarily, the training set can be represented by T, T = {(s1, y1)1, (s2, y2)2,..., (s w , y w ) w ,} = {(s i , y i ) i , 1 ≤ i ≤ w};

[0037] The validation set can be represented by E, E = {(s1, y1)1, (s2, y2)2,..., (s v , y v ) v ,} = {(s i , y i ) i , 1 ≤ i ≤ v};

[0038] Among them, s i represents the sample data of the i-th sample (i.e., the description information of the item), y i represents the sample label of the i-th sample (i.e., the standard category to which the item belongs in the standard category system), w represents the number of samples in the training set, and v represents the number of samples in the validation set.

[0039] In the model training stage, the sample data s i in the training set can be used as the model training input to obtain the training output of the model. According to the training output of the model and the corresponding sample label y iDetermine the loss function, and reverse-optimize the model parameters based on the loss function to obtain a category recognition model. In the model verification stage, the performance of the category recognition model obtained in the training stage can be evaluated according to the validation set. If the performance meets the requirements, the trained category recognition model is put into use. If the performance does not meet the requirements, continue iterative training until a category recognition model with performance meeting the requirements is obtained.

[0040] Specifically, in the embodiments of the present invention, as Figure 2 shown, the trained category recognition model may include three layers: an input layer, a mapping layer, and a decoding layer. Using the category recognition model to identify the second standard category to which the item to be recognized belongs in the standard category system may include:

[0041] (1) Use the input layer to segment the description information into a word set, and determine the corresponding original vector sequence of the word set based on the bag of words.

[0042] When the item to be recognized is an industrial product, the input layer can be used to perform industrial word segmentation on the description information of the item to be recognized to obtain a word set. Specifically, the industrial word segmentation can be implemented by at least one of the following three methods: perform industrial word segmentation based on a pre-established industrial word library, perform industrial word segmentation based on a pre-established industrial word segmentation rule, and perform industrial word segmentation using a pre-established industrial word segmentation model. Performing industrial word segmentation on the description information of the item to be recognized can improve the accuracy of word segmentation and the accuracy of subsequent commodity category recognition.

[0043] Specifically, the bag of words can be established by performing industrial word segmentation on the sample data in the training set and the validation set. The bag of words can be used to map the position (such as position id) of each word in the word set obtained by word segmentation, and construct a word vector for each word according to each word position, so as to obtain the corresponding original vector sequence of the word set. For example, if there are 5 words in the word set, 5 word vectors will be obtained, and these 5 word vectors form the original vector sequence. Since the number of words in the bag of words is large, each word vector in the original vector sequence is a high-dimensional vector.

[0044] (2) Use the mapping layer to perform dimensionality reduction processing on the original vector sequence to obtain a dimensionality-reduced vector sequence, and perform fusion processing on the dimensionality-reduced vector sequence to obtain a fusion vector.

[0045] The dimensionality-reduced vector sequence obtained by performing dimensionality reduction on the original vector sequence can represent the semantic information in the description information of the item to be recognized, thereby improving the accuracy of subsequent category recognition. After obtaining the dimensionality-reduced vector sequence, a fusion process can be performed on the dimensionality-reduced vector sequence. For example, the vectors in the dimensionality-reduced vector sequence can be summed and averaged to obtain a vector, that is, the fusion vector. For example, if the original vector sequence includes 5 high-dimensional (e.g., 10,000-dimensional) vectors, these 5 high-dimensional vectors can be dimensionally reduced to obtain 5 low-dimensional (e.g., 300-dimensional) vectors, and then the 5 low-dimensional vectors are summed and averaged to obtain a low-dimensional (e.g., 300-dimensional) vector, that is, the fusion vector.

[0046] (3) Use the decoding layer to decode and classify the fusion vector to obtain the second standard category.

[0047] Exemplarily, the decoding layer can be composed of a logical classification model. For example, a Huffman tree. When performing decoding and classification, for the fusion vector output by the mapping layer, binary classification logistic regression can be performed on each node in the path according to the position of the leaf node of the item corresponding category in the Huffman tree, so as to obtain the standard category to which the item to be recognized belongs, that is, the second standard category. Among them, the Huffman tree can be constructed by organizing each category according to the number of items in each category. Using the Huffman tree for decoding and classification can improve the classification efficiency.

[0048] Using the above three-layer category recognition model to recognize the standard category to which the item to be recognized belongs in the standard category system can improve the recognition efficiency and accuracy.

[0049] Step 104, determine the target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and / or the second standard category.

[0050] In practical applications, after obtaining the first standard category, the score corresponding to the first standard category can be obtained. If the score corresponding to the first standard category meets the conditions, step 103 can be skipped and the first standard category can be determined as the target standard category; if the score corresponding to the first standard category does not meet the conditions, step 103 is executed to obtain the second standard category, and the second standard category is determined as the target standard category. Specifically, before determining the second standard category as the target standard category, the score corresponding to the second standard category can also be obtained. If the score corresponding to the second standard category meets the conditions, the second standard category is determined as the target standard category; if the score corresponding to the second standard category does not meet the conditions, category recognition failure can be feedback, or it can be prompted for manual category recognition. That is, using the model to recognize categories as a supplement to the category system mapping to recognize categories, making up for the problem of incomplete recognition of categories by the category system mapping, reducing the manual intervention in category recognition, and improving the recognition probability and recognition accuracy.

[0051] Specifically, the category system mapping recognition category and the model recognition category can also be adopted simultaneously. After obtaining the first standard category and the second standard category, obtain the score corresponding to the first standard category, and obtain the score corresponding to the second standard category. Select the category with the higher score from the first standard category and the second standard category, and use the category with the higher score as the target standard category.

[0052] In the embodiments of the present invention, the method of category system mapping recognition category and the method of model recognition category are combined to realize the category recognition of items. The entire category recognition method can be implemented by machine mounting automatically, with high recognition efficiency and reduced labor costs. In addition, by combining the method of category system mapping recognition category and the method of model recognition category, the accuracy of category recognition is improved, and more fine-grained category recognition can be realized.

[0053] The method of using category system mapping to recognize categories is further introduced below. As Figure 3 shown, that is, Figure 1 step 102 in

[0054] Step 1021: Query the preset mapping relationship based on the original category to obtain multiple candidate standard categories to which the item to be recognized belongs in the standard category system.

[0055] The preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the preset mapping relationship can be established by manually or machine analyzing the original category system and the standard category system. Each candidate standard category in the multiple candidate standard categories can be represented by the corresponding terminal category in the standard category system, or can be represented by the corresponding complete level category in the standard category system.

[0056] Step 1022: Obtain the keywords corresponding to each candidate standard category.

[0057] The keywords corresponding to each candidate standard category can be preset manually or by machine.

[0058] Step 1023: Calculate the character similarity between the keywords corresponding to each candidate standard category and the description information of the item to be recognized, so as to obtain the correlation score between each candidate standard category and the description information.

[0059] In specific implementation, the keywords corresponding to multiple candidate standard categories can form a set. For example, denote this set as P, then P = {p1, p2,..., p n ,} = {p i , 1 ≤ i ≤ n}, p iLet \(P_i\) denote the keyword corresponding to the \(i\)-th candidate standard category, and \(n\) denote the number of candidate standard categories. Suppose the description information of the item to be recognized is \(q\), that is, it is necessary to find the keyword in \(P\) that is most relevant to \(q\), which can be found by calculating the character similarity. The specific calculation formula is as follows:

[0060]

[0061] where \(CTR\) i denotes the character similarity between \(q\) and the \(i\)-th keyword in \(P\), \(L(x)\) denotes the number of characters in the information \(x\), \(set(x)\) denotes the set of non-repeating characters in the information \(x\), and \(&\) denotes the intersection of two sets.

[0062] After obtaining the character similarity between each keyword in \(P\) and \(q\), the character similarity between each keyword in \(P\) and \(q\) can be determined as the relevance score between the corresponding candidate standard category and the description information of the item to be recognized.

[0063] Step 1024: Select the candidate standard category with the highest relevance score from multiple candidate standard categories to obtain the first standard category.

[0064] In this embodiment, when using the category system mapping to identify categories, the first standard category is determined based on the character similarity, and the algorithm is simple and has high accuracy.

[0065] Next, a method for identifying categories using a model is further introduced. In the embodiments of the present invention, the category identification model can include two types: a full-level category identification model and a non-full-level category identification model. Among them, the category identification granularity of the full-level category identification model is finer than that of the non-full-level category identification model. The full-level category identification model is trained based on the first sample set, and the non-full-level category identification model is trained based on the second sample set. The sample labels in the first sample set are full-level category labels, and the sample labels in the second sample set are non-full-level category labels. For example, the sample labels in the first sample set can be the terminal categories of the standard category system, and the sample labels in the second sample set are the sub-terminal categories, secondary categories, or primary categories, etc. of the standard category system. That is, the full-level category identification model can identify the terminal category to which the item to be recognized belongs, and the non-full-level category identification model can only identify the sub-terminal category, secondary category, or primary category, etc. to which the item to be recognized belongs.

[0066] As Figure 4 shown, that is Figure 1 Step 103 in can include the following steps:

[0067] Step 1031: Input the description information into the full-level category identification model for category identification to obtain the direct full-level category to which the item to be recognized belongs under the standard category system.

[0068] Specifically, the description information of the item to be recognized can be input into the full-level category recognition model, and the output of the full-level category recognition model is the direct full-level category to which the item to be recognized belongs in the standard category system and the confidence level of this direct full-level category. That is, the full-level category recognition model can be used to identify the last-level category to which the item to be recognized belongs in the standard category system.

[0069] Step 1032: Determine whether the confidence level of the direct full-level category exceeds the preset confidence threshold. If it exceeds, execute Step 1039; otherwise, execute Step 1033.

[0070] The preset threshold, such as 0.8, 0.9, etc. Exceeding the preset confidence threshold indicates that the recognized last-level category has a relatively high credibility and can be output as the category recognition result.

[0071] Step 1033: Input the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs in the standard category system.

[0072] For example, the description information of the item to be recognized can be input into the non-full-level category recognition model, and the output of the non-full-level category recognition model is the non-full-level category to which the item to be recognized belongs in the standard category system, such as the penultimate-level category, secondary category, primary category, etc. The specific level of the output category is determined by the model training process.

[0073] Step 1034: Based on the non-full-level category, reason and recall multiple candidate full-level categories to which the item to be recognized belongs in the standard category system.

[0074] That is, based on the penultimate-level category, secondary category, primary category, etc., reason out the last-level categories that the item to be recognized may belong to in the standard category system.

[0075] Step 1035: Obtain the keywords corresponding to each candidate full-level category.

[0076] The keywords corresponding to each candidate full-level category can be preset manually or by machine.

[0077] Step 1036: Calculate the character similarity between the keywords corresponding to each candidate full-level category and the description information, so as to obtain the correlation score between each candidate full-level category and the description information.

[0078] For the specific calculation method, please refer to the description of the previous embodiments and will not be elaborated here.

[0079] Step 1037: Select the candidate full-level category with the highest correlation score from multiple candidate full-level categories to obtain the inferred full-level category.

[0080] Step 1038: Determine the inferred full-level category as the second standard category.

[0081] Step 1039, determine the direct full-level category as the second standard category.

[0082] In specific implementation, there may be multiple non-full-level category recognition models. Taking the example that the multiple non-full-level category recognition models include a first non-full-level category recognition model and a second non-full-level category recognition model, if the category recognition granularity of the first non-full-level category recognition model is finer than that of the second non-full-level category recognition model, then input the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs in the standard category system, and determine the inferred full-level category to which the item to be recognized belongs in the standard category system according to the non-full-level category. The steps may include the following:

[0083] (1) Input the description information into the first non-full-level category recognition model for category recognition to obtain the first non-full-level category to which the item to be recognized belongs in the standard category system, and determine the first inferred full-level category to which the item to be recognized belongs in the standard category system according to the first non-full-level category.

[0084] Among them, for the full-level category to which the item to be recognized belongs in the standard category system determined according to the first non-full-level category, there are still multiple ones. It is necessary to determine the first inferred full-level category from the multiple full-level categories based on the relevance scores of each full-level category in the multiple full-level categories with the description information for recognizing the item to be recognized.

[0085] (2) Determine the relevance score between the first inferred full-level category and the description information.

[0086] (3) When the relevance score between the first inferred full-level category and the description information exceeds the preset score threshold, determine the first inferred full-level category as the inferred full-level category.

[0087] (4) When the relevance score between the first inferred full-level category and the description information does not exceed the preset score threshold, input the description information into the second non-full-level category recognition model for category recognition to obtain the second non-full-level category to which the item to be recognized belongs in the standard category system, and determine the second inferred full-level category to which the item to be recognized belongs in the standard category system according to the second non-full-level category, and determine the second inferred full-level category as the inferred full-level category.

[0088] That is, when there are multiple non-full-level category recognition models, preferentially use the non-full-level category recognition model with finer recognition granularity. If the recognition result of the non-full-level category recognition model with finer recognition granularity does not meet the requirements, then enable the non-full-level category recognition model with coarser recognition granularity. Of course, in actual application, multiple non-full-level category recognition models can also be enabled simultaneously, or the full-level category recognition model and multiple non-full-level category recognition models can be enabled simultaneously, and the output with the best recognition result (such as the highest confidence) is taken.

[0089] In this embodiment, two types of category recognition models (full-level category recognition model and non-full-level category recognition model) are trained to make up for the situation where a model with a relatively coarse recognition granularity fails to recognize results or recognizes inaccurately, reduce the manual intervention in category recognition, and improve the recognition probability and accuracy.

[0090] The following further describes the item category recognition method provided by the embodiments of the present invention. As Figure 5 shown, it may include the following steps:

[0091] Step 301, obtain the description information of the item to be recognized, and obtain the original category to which the item to be recognized belongs in the original category system.

[0092] Step 302, query the preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs in the standard category system.

[0093] Step 303, determine whether the correlation score between the first standard category and the description information exceeds the preset score threshold. If it exceeds, execute Step 306; otherwise, execute Step 304.

[0094] For the calculation of the correlation score between the first standard category and the description information, reference can be made to the description of the previous embodiments, which will not be elaborated here. The preset score threshold can be set according to actual needs or experience.

[0095] Step 304, input the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs in the standard category system.

[0096] Step 305, determine the second standard category as the target standard category.

[0097] Specifically, before determining the second standard category as the target standard category, the confidence level corresponding to the second standard category can also be obtained. If the confidence level corresponding to the second standard category exceeds the preset confidence level threshold, the second standard category is determined as the target standard category; if the confidence level corresponding to the second standard category does not exceed the preset confidence level threshold, category recognition failure can be fed back, or it can be prompted to perform category recognition manually.

[0098] Step 306, determine the first standard category as the target standard category.

[0099] In the embodiments of the present invention, the method of mapping and identifying categories in the category system is combined with the method of identifying categories by the model to realize the category identification of items. The entire category identification method can be implemented by machine mounting automatically, with high identification efficiency and reduced labor costs. In addition, using the model to identify categories as a supplement to the method of mapping and identifying categories in the category system can make up for the problem of incomplete identification of categories in the category system mapping method, reduce the manual intervention in category identification, and improve the identification probability and accuracy.

[0100] The following takes a standard category system including four levels of categories: primary categories, secondary categories, sub-final categories, and final categories, and a category identification model including a full-level category identification model (such as Figure 6 Model 1 in, and Model 1 can identify the final category) and a non-full-level category identification model (such as Figure 6 Model 2, Model 3, and Model 4 in, Model 2 can identify the sub-final category, Model 3 can identify the secondary category, and Model 4 can identify the primary category) as an example to illustrate the item category identification method provided by the embodiments of the present invention. As Figure 6 shown, after obtaining the description information of the item to be identified and the original category to which the item to be identified belongs in the original category system, first use the method of mapping and identifying categories in the category system, that is, Route ①. Based on the original category, recall multiple standard categories to which the item to be identified belongs in the standard category system using the preset mapping relationship to obtain a standard category set. Perform fine ranking on the standard categories in the standard category set by calculating the correlation scores, select the standard category with the highest correlation score according to the fine ranking result, and determine whether the correlation score of the selected standard category exceeds the preset score threshold. If it exceeds, it meets the exemption condition, and output the standard category as the target standard category.

[0101] If the standard category obtained by Route ① does not exceed the preset score threshold, that is, it does not meet the exemption condition, then use the model for identification. When using the model for identification, first execute Route ②, use Model 1 for identification to obtain the direct final category to which the item to be identified belongs (the full-level category can be directly obtained based on the direct final category); determine whether the confidence level corresponding to the direct final category exceeds the preset confidence level threshold. If it exceeds, it meets the exemption condition, and output the direct final category as the target standard category.

[0102] If the direct final category obtained by Route ② does not exceed the preset confidence level threshold, that is, it does not meet the exemption condition, then execute Route ③, use Model 2 for identification to obtain the sub-final category to which the item to be identified belongs, infer multiple inferred final categories based on the sub-final category, perform fine ranking on the multiple inferred final categories by calculating the correlation scores, select the inferred final category with the highest correlation score according to the fine ranking result, and determine whether the correlation score of the selected inferred final category exceeds the preset score threshold. If it exceeds, it meets the exemption condition, and output the inferred final category as the target standard category.

[0103] If the relevance score of the inferred terminal category obtained from line ③ does not exceed the preset score threshold, the non-exemption condition is not met, and then line ④ is executed. The penultimate category is identified using Model 3, and the subsequent execution method is similar to that of line ③. If the inferred terminal category obtained after line ④ meets the exemption condition, the inferred terminal category obtained from line ④ is output as the target standard category.

[0104] If the inferred terminal category obtained after line ④ does not meet the exemption condition, line ⑤ is executed. The primary category is identified using Model 4, and the subsequent execution method of line ⑤ is similar to that of line ③. If the inferred terminal category obtained after line ⑤ meets the exemption condition, the inferred terminal category obtained from line ⑤ is output as the target standard category.

[0105] If the inferred terminal category obtained after line ⑤ does not meet the exemption condition, it is possible to feedback that the category identification fails, or prompt for manual category identification.

[0106] It should be noted that the execution order of the lines in the above example is ①②③④⑤, and this execution order is a relatively reasonable one obtained through verification. In actual applications, the execution order of the lines can also be adaptively adjusted according to business needs or scenarios. As long as the overall idea is to combine the method of mapping the category system to identify categories and the method of using models to identify categories, it is within the protection scope of the present invention.

[0107] Figure 7 is a structural diagram of an article category identification device provided by an embodiment of the present invention. This device is applicable to execute the article category identification method provided by the embodiment of the present invention. As Figure 7 shown, the device may specifically include:

[0108] An acquisition module 401, configured to acquire the description information of the item to be identified, and acquire the original category to which the item to be identified belongs in the original category system;

[0109] A query module 402, configured to query a preset mapping relationship based on the original category to obtain a first standard category to which the item to be identified belongs in the standard category system. The preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category grading granularity in the standard category system is finer than that in the original category system;

[0110] An identification module 403, configured to input the description information into a category identification model for category identification to obtain a second standard category to which the item to be identified belongs in the standard category system;

[0111] A determination module 404, configured to determine a target standard category to which the item to be identified belongs under the standard category system based on the first standard category and / or the second standard category.

[0112] In one embodiment, the query module 402 is specifically configured to:

[0113] Query a preset mapping relationship based on the original category to obtain multiple candidate standard categories to which the item to be identified belongs under the standard category system;

[0114] Calculate a relevance score between each candidate standard category in the multiple candidate standard categories and the description information;

[0115] Determine the first standard category from the multiple candidate standard categories based on the relevance score.

[0116] In one embodiment, when the query module 402 calculates the relevance score between each candidate standard category in the multiple candidate standard categories and the description information, it includes:

[0117] Obtain keywords corresponding to each candidate standard category;

[0118] Calculate the character similarity between the keywords corresponding to each candidate standard category and the description information, so as to obtain the relevance score between each candidate standard category and the description information.

[0119] In one embodiment, when the query module 402 determines the first standard category from the multiple candidate standard categories based on the relevance score, it includes:

[0120] Select the candidate standard category with the highest relevance score from the multiple candidate standard categories to obtain the first standard category.

[0121] In one embodiment, the recognition module 403 is specifically configured to:

[0122] Determine whether the relevance score between the first standard category and the description information exceeds a preset score threshold;

[0123] When the relevance score between the first standard category and the description information does not exceed the preset score threshold, perform inputting the description information into a category recognition model for category recognition to obtain a second standard category to which the item to be identified belongs under the standard category system.

[0124] In one embodiment, the determination module 404 is specifically configured to:

[0125] When the relevance score between the first standard category and the description information exceeds the preset score threshold, determine the first standard category as the target standard category.

[0126] In one embodiment, the determining module 404 is specifically configured to:

[0127] When the relevance score between the first standard category and the description information does not exceed the preset score threshold, determine the second standard category as the target standard category.

[0128] In one embodiment, the category recognition model includes a full-level category recognition model and a non-full-level category recognition model. The category recognition granularity of the full-level category recognition model is finer than that of the non-full-level category recognition model. The recognition module 403 is specifically configured to:

[0129] Input the description information into the full-level category recognition model for category recognition to obtain the direct full-level category to which the item to be recognized belongs in the standard category system;

[0130] Input the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs in the standard category system, and determine the inferred full-level category to which the item to be recognized belongs in the standard category system according to the non-full-level category;

[0131] Based on the direct full-level category and / or the inferred full-level category, determine the second standard category to which the item to be recognized belongs in the standard category system.

[0132] In one embodiment, the full-level category recognition model is trained based on a first sample set, and the non-full-level category recognition model is trained based on a second sample set. The sample labels in the first sample set are full-level category labels, and the sample labels in the second sample set are non-full-level category labels.

[0133] In one embodiment, before inputting the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs in the standard category system, the recognition module 403 is further configured to:

[0134] Determine whether the confidence level of the direct full-level category exceeds a preset confidence level threshold;

[0135] When the confidence level of the direct full-level category does not exceed the preset confidence level threshold, perform the step of inputting the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs in the standard category system.

[0136] In one embodiment, the recognition module 403 determines the second standard category to which the item to be recognized belongs in the standard category system based on the direct full-level category and / or the inferred full-level category, including:

[0137] When the confidence level of the direct full - level category exceeds the preset confidence threshold, determine the direct full - level category as the second standard category.

[0138] In one embodiment, the recognition module 403 determines the second standard category to which the item to be recognized belongs under the standard category system based on the direct full - level category and / or the inferred full - level category, including:

[0139] When the confidence level of the direct full - level category does not exceed the preset confidence threshold, determine the inferred full - level category as the second standard category.

[0140] In one embodiment, the recognition module 403 determines the inferred full - level category to which the item to be recognized belongs under the standard category system according to the non - full - level category, including:

[0141] Recall multiple candidate full - level categories to which the item to be recognized belongs under the standard category system based on the non - full - level category;

[0142] Calculate the correlation score between each candidate full - level category in the multiple candidate full - level categories and the description information;

[0143] Determine the inferred full - level category from the multiple candidate full - level categories based on the correlation score.

[0144] In one embodiment, the recognition module 403 calculates the correlation score between each candidate full - level category in the multiple candidate full - level categories and the description information, including:

[0145] Obtain the keywords corresponding to each candidate full - level category;

[0146] Calculate the character similarity between the keywords corresponding to each candidate full - level category and the description information, so as to obtain the correlation score between each candidate full - level category and the description information.

[0147] In one embodiment, the recognition module 403 determines the inferred full - level category from the multiple candidate full - level categories based on the correlation score, including:

[0148] Select the candidate full - level category with the highest correlation score from the multiple candidate full - level categories to obtain the inferred full - level category.

[0149] In one embodiment, the non-full-level category recognition model includes a first non-full-level category recognition model and a second non-full-level category recognition model. The category recognition granularity of the first non-full-level category recognition model is finer than that of the second non-full-level category recognition model. The recognition module 403 inputs the description information into the non-full-level category recognition model for category recognition, obtains the non-full-level category to which the item to be recognized belongs in the standard category system, and determines the inferred full-level category to which the item to be recognized belongs in the standard category system according to the non-full-level category, including:

[0150] Input the description information into the first non-full-level category recognition model for category recognition, obtain the first non-full-level category to which the item to be recognized belongs in the standard category system, and determine the first inferred full-level category to which the item to be recognized belongs in the standard category system according to the first non-full-level category;

[0151] Determine the relevance score between the first inferred full-level category and the description information;

[0152] When the relevance score between the first inferred full-level category and the description information exceeds the preset score threshold, determine the first inferred full-level category as the inferred full-level category;

[0153] When the relevance score between the first inferred full-level category and the description information does not exceed the preset score threshold, input the description information into the second non-full-level category recognition model for category recognition, obtain the second non-full-level category to which the item to be recognized belongs in the standard category system, and determine the second inferred full-level category to which the item to be recognized belongs in the standard category system according to the second non-full-level category, and determine the second inferred full-level category as the inferred full-level category.

[0154] In one embodiment, the category recognition model includes an input layer, a mapping layer, and a decoding layer. The recognition module 403 is specifically used for:

[0155] Use the input layer to segment the description information into a word set, and determine the corresponding original vector sequence of the word set based on the bag of words;

[0156] Use the mapping layer to perform dimensionality reduction processing on the original vector sequence to obtain a dimensionality-reduced vector sequence, and perform fusion processing on the dimensionality-reduced vector sequence to obtain a fusion vector;

[0157] Use the decoding layer to perform decoding classification on the fusion vector to obtain the first standard category.

[0158] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the division of the above functional modules is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the above-described functional modules can refer to the corresponding process in the foregoing method embodiments and will not be elaborated herein.

[0159] The device according to an embodiment of the present invention can obtain the description information of the item to be recognized and obtain the original category to which the item to be recognized belongs in the original category system; query the preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than that in the original category system; input the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs in the standard category system; determine the target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and / or the second standard category. That is, in the embodiment of the present invention, the method of mapping the category system to recognize categories and the method of using the model to recognize categories are combined to realize the category recognition of items. The entire category recognition method can be implemented by machine mounting automatically, with high recognition efficiency and reduced labor costs; in addition, by combining the method of mapping the category system to recognize categories and the method of using the model to recognize categories, the accuracy of category recognition is improved, and more fine-grained category recognition can be achieved.

[0160] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the item category recognition method provided in any one of the above embodiments.

[0161] An embodiment of the present invention also provides a computer-readable medium, on which a computer program is stored. When the program is executed by a processor, it implements the item category recognition method provided in any one of the above embodiments.

[0162] Next, refer to Figure 8 , which shows a schematic structural diagram of a computer system 500 of an electronic device suitable for implementing the embodiment of the present invention. Figure 8 The shown electronic device is only an example and should not impose any limitations on the functions and usage scope of the embodiment of the present invention.

[0163] As Figure 8As shown, computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage section 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0164] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed so that a computer program read from it can be installed into the storage section 508 as needed.

[0165] Specifically, according to an embodiment disclosed by the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment disclosed by the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 509, and / or installed from the removable medium 511. When the computer program is executed by a central processing unit (CPU) 501, the above functions defined in the system of the present invention are executed.

[0166] It should be noted that the computer-readable medium shown in the present invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present invention, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which the computer-readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0167] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the above module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0168] The modules and / or units involved in the embodiments of the present invention can be implemented in software or in hardware. The described modules and / or units can also be provided in a processor. For example, it can be described as: a processor includes a gradient compression module, a gradient transmission module, a momentum determination module, and a model update module; or, it can be described as: a processor includes an acquisition module, a query module, an identification module, and a determination module. Among them, the names of these modules do not constitute a limitation to the module itself in some cases.

[0169] As another aspect, the present invention also provides a computer-readable medium, which can be included in the device described in the above embodiments; or it can exist alone without being assembled into the device. The above computer-readable medium carries one or more programs. When the one or more programs are executed by the device, the device includes: obtaining the description information of the item to be identified, and obtaining the original category to which the item to be identified belongs in the original category system; querying a preset mapping relationship based on the original category to obtain the first standard category to which the item to be identified belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than the category classification granularity in the original category system; inputting the description information into a category identification model for category identification to obtain the second standard category to which the item to be identified belongs in the standard category system; determining the target standard category to which the item to be identified belongs in the standard category system based on the first standard category and / or the second standard category.

[0170] According to the technical solution of the embodiments of the present invention, the method of mapping the category system to identify categories and the method of using a model to identify categories can be combined to implement the category identification of items. The entire category identification method can be implemented by machine mounting automatically, with high identification efficiency and reduced labor costs. In addition, by combining the method of mapping the category system to identify categories and the method of using a model to identify categories, the accuracy of category identification is improved, and more fine-grained category identification can be achieved.

[0171] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An item category recognition method, characterized in that, Including: Obtain the description information of the item to be recognized, and obtain the original category to which the item to be recognized belongs in the original category system; Query the preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs in the standard category system. The preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than that in the original category system; There are multiple standard categories to which the item to be recognized belongs in the standard category system recalled based on the original category to which the item to be recognized belongs by using the preset mapping relationship. The first standard category is the standard category selected from multiple standard categories; Input the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs in the standard category system; The sample data used to train the category recognition model is the description information of the item, and the sample label is the standard category to which the corresponding item belongs in the standard category system; Determine the target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and the second standard category.

2. The method according to claim 1, characterized in that, The step of querying the preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs in the standard category system includes: Query the preset mapping relationship based on the original category to obtain multiple candidate standard categories to which the item to be recognized belongs in the standard category system; Calculate the correlation score between each candidate standard category in the multiple candidate standard categories and the description information; Determine the first standard category from the multiple candidate standard categories based on the correlation score.

3. The method according to claim 2, wherein The step of calculating the correlation score between each candidate standard category in the multiple candidate standard categories and the description information includes: Obtain the keywords corresponding to each candidate standard category; Calculate the character similarity between the keywords corresponding to each candidate standard category and the description information, so as to obtain the correlation score between each candidate standard category and the description information.

4. The method according to claim 2, wherein The step of determining the first standard category from the multiple candidate standard categories based on the correlation score includes: Select the candidate standard category with the highest correlation score from the multiple candidate standard categories to obtain the first standard category.

5. The method according to claim 2, wherein Before inputting the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs in the standard category system, it further includes: Determine whether the correlation score between the first standard category and the description information exceeds the preset score threshold; When the correlation score between the first standard category and the description information does not exceed the preset score threshold, perform the step of inputting the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs in the standard category system.

6. The method according to claim 5, wherein The step of determining the target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and the second standard category includes: When the relevance score between the first standard category and the description information exceeds the preset score threshold, determine the first standard category as the target standard category.

7. The method according to claim 5, wherein The determining the target standard category to which the item to be recognized belongs under the standard category system based on the first standard category and the second standard category includes: When the relevance score between the first standard category and the description information does not exceed the preset score threshold, determine the second standard category as the target standard category.

8. The method according to claim 1, wherein The category recognition model includes a full-level category recognition model and a non-full-level category recognition model. The category recognition granularity of the full-level category recognition model is finer than that of the non-full-level category recognition model. The inputting the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs under the standard category system includes: Input the description information into the full-level category recognition model for category recognition to obtain the direct full-level category to which the item to be recognized belongs under the standard category system; Input the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs under the standard category system, and determine the inferred full-level category to which the item to be recognized belongs under the standard category system according to the non-full-level category; Based on the direct full-level category and the inferred full-level category, determine the second standard category to which the item to be recognized belongs under the standard category system.

9. The method according to claim 8, wherein The full-level category recognition model is trained based on a first sample set, and the non-full-level category recognition model is trained based on a second sample set. The sample labels in the first sample set are full-level category labels, and the sample labels in the second sample set are non-full-level category labels.

10. The method according to claim 8, wherein Before inputting the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs under the standard category system, it further includes: Determine whether the confidence level of the direct full-level category exceeds the preset confidence level threshold; When the confidence level of the direct full-level category does not exceed the preset confidence level threshold, perform the step of inputting the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs under the standard category system.

11. The method according to claim 10, wherein The determining the second standard category to which the item to be recognized belongs under the standard category system based on the direct full-level category and the inferred full-level category includes: When the confidence level of the direct full-level category exceeds the preset confidence level threshold, determine the direct full-level category as the second standard category.

12. The method according to claim 10, characterized in that, characterized in that, The determining the second standard category to which the item to be recognized belongs under the standard category system based on the direct full-level category and the inferred full-level category includes: When the confidence level of the direct full-level category does not exceed the preset confidence level threshold, determine the inferred full-level category as the second standard category.

13. The method according to claim 8, wherein The determining the inferred full-level category to which the item to be recognized belongs under the standard category system according to the non-full-level category includes: Recall multiple candidate full-level categories to which the item to be recognized belongs under the standard category system based on the non-full-level category; Calculate the relevance score between each candidate full-level category in the multiple candidate full-level categories and the description information; Determine the inferred full-level category from the multiple candidate full-level categories based on the relevance score.

14. The method according to claim 13, wherein The calculating the relevance score between each candidate full-level category in the multiple candidate full-level categories and the description information includes: Obtain the keywords corresponding to each candidate full-level category; Calculate the character similarity between the keywords corresponding to each candidate full-level category and the description information, so as to obtain the relevance score between each candidate full-level category and the description information.

15. The method according to claim 13, characterized in that, The determining the inferred full-level category from the multiple candidate full-level categories based on the relevance score includes: Select the candidate full-level category with the highest relevance score from the multiple candidate full-level categories to obtain the inferred full-level category.

16. The method according to claim 8, characterized in that, The non-full-level category recognition model includes a first non-full-level category recognition model and a second non-full-level category recognition model. The category recognition granularity of the first non-full-level category recognition model is finer than that of the second non-full-level category recognition model. The inputting the description information into the non-full-level category recognition model for category recognition to obtain the non-full-level category to which the item to be recognized belongs under the standard category system, and determining the inferred full-level category to which the item to be recognized belongs under the standard category system according to the non-full-level category includes: Input the description information into the first non-full-level category recognition model for category recognition to obtain the first non-full-level category to which the item to be recognized belongs under the standard category system, and determine the first inferred full-level category to which the item to be recognized belongs under the standard category system according to the first non-full-level category; Determine the relevance score between the first inferred full-level category and the description information; When the relevance score between the first inferred full-level category and the description information exceeds the preset score threshold, determine the first inferred full-level category as the inferred full-level category; When the relevance score between the first inferred full-level category and the description information does not exceed the preset score threshold, input the description information into the second non-full-level category recognition model for category recognition to obtain the second non-full-level category to which the item to be recognized belongs under the standard category system, and determine the second inferred full-level category to which the item to be recognized belongs under the standard category system according to the second non-full-level category, and determine the second inferred full-level category as the inferred full-level category.

17. The method according to claim 1, characterized in that, The category recognition model includes an input layer, a mapping layer and a decoding layer. The inputting the description information into the category recognition model for category recognition to obtain the second standard category to which the item to be recognized belongs under the standard category system includes: Use the input layer to segment the description information into a word set, and determine the corresponding original vector sequence of the word set based on the bag of words; The original vector sequence is dimensionally reduced using the mapping layer to obtain a dimensionally reduced vector sequence, and the dimensionally reduced vector sequence is fusion-processed to obtain a fusion vector; The fusion vector is decoded and classified using the decoding layer to obtain the second standard category.

18. An article category recognition device, characterized in that, It includes: An acquisition module, configured to acquire the description information of the item to be recognized, and acquire the original category to which the item to be recognized belongs in the original category system; A query module, configured to query a preset mapping relationship based on the original category to obtain the first standard category to which the item to be recognized belongs in the standard category system, where the preset mapping relationship includes the mapping relationship between the original category system and the standard category system, and the category classification granularity in the standard category system is finer than the category classification granularity in the original category system; When recalling the standard category to which the item to be recognized belongs in the standard category system based on the original category to which the item to be recognized belongs using the preset mapping relationship, there are multiple standard categories, and the first standard category is the standard category selected from the multiple standard categories; An identification module, configured to input the description information into a category identification model for category identification to obtain the second standard category to which the item to be recognized belongs in the standard category system; The sample data used for training the category identification model is the description information of the item, and the sample label is the standard category to which the corresponding item belongs in the standard category system; A determination module, configured to determine the target standard category to which the item to be recognized belongs in the standard category system based on the first standard category and the second standard category.

19. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the item category identification method according to any one of claims 1 to 17.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the item category identification method according to any one of claims 1 to 17.

Images