Method and apparatus for determining goods tariff code based on deep learning

Abstract

The present disclosure relates to a method and apparatus for determining a goods tariff code based on deep learning. The method comprises: acquiring declaration data of declared goods; on the basis of the declaration data, determining a sub-category, for tariff code classification, of the declared goods; and determining a predicted tariff code of the declared goods on the basis of the determined sub-category. In embodiments of the present disclosure, the method can further comprise comparing the predicted tariff code of the declared goods with a declared tariff code in the declaration data to determine whether the current declaration is abnormal. A deep learning model according to the embodiments of the present disclosure can automatically predict the tariff code of the declared goods and automatically complete tariff code abnormal declaration detection.

Description

Method and device for determining commodity tax number based on deep learning

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority from the Chinese patent application number 202311622667.0 filed on November 29, 2023, and entitled “Method and device for determining commodity tax numbers based on deep learning,” and all of its contents are incorporated herein. Technical Field

[0003] The present disclosure generally relates to the field of deep learning, and in particular, to a method and apparatus for determining commodity tax numbers based on deep learning. Background Art

[0004] Abnormal customs declarations of commodity tax numbers refer to errors, omissions, or deliberate misrepresentation of commodity tax numbers by importers and exporters in international trade. Currently, customs verification of commodity tax numbers relies primarily on manual review and rule-based engines. However, due to the wide variety and complex characteristics of commodities, manual review is time-consuming, labor-intensive, and inefficient. Furthermore, traditional rule-based engines use predefined rule sets to detect abnormal tax number declarations. These rules, based on empirical knowledge and manual definitions, cannot cover complex anomalies and require constant updating and maintenance of rule sets to adapt to new trade patterns.

[0005] Abnormal tax number declarations can lead to losses in customs revenue, disrupt market order, and even involve illegal activities such as smuggling and tax evasion. To ensure fair trade and effective tax collection, customs authorities need effective methods to detect and correct abnormal tax number declarations for customs goods. Summary of the Invention

[0006] According to a first aspect of the present disclosure, a method for determining a commodity tax number is provided, comprising: obtaining declaration data of a declared commodity; determining a subcategory of the declared commodity for tax number classification based on the declaration data; and determining a predicted tax number of the declared commodity based on the subcategory.

[0007] According to a second aspect of the present disclosure, a device for determining a commodity tax number is provided, comprising: a processor; and a memory storing instructions that, when executed by the processor, implement the method according to the first aspect of the present disclosure.

[0008] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which instructions are stored. When the instructions are executed, the method according to the first aspect of the present disclosure is implemented.

[0009] According to a fourth aspect of the present disclosure, a device for determining a commodity tax number is provided, wherein the device includes a module for executing the method according to the first aspect of the present disclosure.

[0010] This disclosure proposes a method for determining customs commodity tax numbers based on deep learning. By comparing the tax number determined for a declared commodity with the actual declared tax number of the declared commodity, it is possible to detect abnormal tax number declarations for customs commodities. The deep learning model according to the embodiments of this disclosure can automatically predict the tax number of declared customs commodities and automatically complete the detection of abnormal tax number declarations, thereby reducing the workload of customs departments, improving the efficiency and accuracy of supervision, and thus maintaining fair trade and the stability of tax collection. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The features and advantages of the present disclosure will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present disclosure in any way.

[0012] FIG1 shows a flowchart of a method 100 for determining a commodity tax number according to an embodiment of the present disclosure.

[0013] FIG2 shows a flowchart of a method 200 for determining a subcategory of a commodity for tax code classification according to an embodiment of the present disclosure.

[0014] FIG3 shows a flow chart of a method 300 for determining the wood species of a wood commodity according to an embodiment of the present disclosure.

[0015] FIG4 shows a flowchart of a method 400 for determining a commodity tax number according to an embodiment of the present disclosure.

[0016] FIG5 shows a flowchart of a method 500 for detecting anomalies in tax number declarations for timber commodities according to an embodiment of the present disclosure.

[0017] FIG6 is a schematic block diagram of an apparatus 600 for determining a commodity tax number according to an embodiment of the present disclosure.

[0018] FIG7 illustrates a block diagram of an example computing device in which the apparatus for determining a commodity tax number according to various embodiments of the present disclosure may be used. DETAILED DESCRIPTION

[0019] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present application and are not configured to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0020] This disclosure proposes a method for determining customs commodity tax numbers based on deep learning. Using this method, accurate analysis of declared data enables precise prediction of commodity tax numbers and detection of anomalies in declared tax numbers. This method utilizes deep neural networks to automatically learn key features and models, demonstrating strong generalization capabilities. It can effectively determine commodity tax numbers and detect anomalies in declared tax numbers, providing a highly efficient tool for determining and detecting tax number anomalies.

[0021] The features and exemplary embodiments of various aspects of the present disclosure are described in detail below.In addition, the features, structures, or characteristics described below may be combined in any suitable manner in one or more embodiments.

[0022] FIG1 shows a flow chart of a method 100 for determining a customs commodity tax number according to a disclosed embodiment. As shown in FIG1 , the method 100 includes steps S110 to S130. It should be understood that the method may include more or fewer steps, and this document does not limit this.

[0023] S110: Obtaining declaration data of the declared commodity. The declaration data may be, for example, customs import and export declaration data of the declared customs commodity, which may include at least the declared commodity name, declared tax number, and declared specification elements.

[0024] S120: Determine the subcategory for tax number classification of the declared goods based on the acquired declaration data.

[0025] In one embodiment, the subcategory for tax code classification of the corresponding commodity may be determined based on the declared commodity name and / or the declared specification element information.

[0026] In one embodiment, the subcategories used for tax code classification can be the attributes used in the tariff schedule to classify commodities, such as the commodity type, appearance, processing method, and purpose. In one example, the declared commodity can be a timber commodity, and the subcategories used for tax code classification for timber commodities can be the corresponding log species of the commodity, such as yew logs, red pine logs, etc. In another example, the declared commodity can be a fruit commodity, and the subcategories used for tax code classification for fruit commodities can be the corresponding fruit species of the declared commodity, such as apples, watermelons, or citrus fruits. In yet another example, the declared commodity can be a cotton commodity, and the subcategories used for tax code classification for cotton commodities can be the corresponding processing method of the declared commodity, such as uncombed, combed, carded, or combed.

[0027] In one embodiment, a pre-built prior knowledge base can be used to determine the subcategories of the declared commodities. In one embodiment, a prior knowledge base for one or more categories of customs commodities can be constructed by acquiring datasets related to one or more categories of customs commodities and pre-processing the acquired datasets.

[0028] In one embodiment, the acquired data set may include historical declaration data for one or more categories of customs commodities and / or prior data related to one or more categories of customs commodities. In one embodiment, the historical declaration data for customs commodities may include customs import and export declaration data, where the declaration data includes at least a tax number field, a commodity name field, and a specification element field. In one example, the historical declaration data may include declaration data over a long period of time. In one embodiment, the prior data for customs commodities may include expert knowledge and / or rules related to the names and / or classifications of the corresponding customs commodities, such as data from books and / or the internet. For example, the prior data for timber commodities may include, for example, Latin-related data on timber, including books related to timber and internet data related to the names and / or classifications of timber.

[0029] In one embodiment, the preprocessing of the acquired data set includes field parsing of the acquired original data set. For example, in the historical declaration data of wood commodities, the "Specification Elements" field records the detailed information about the declared commodities that enterprises and / or individuals need to fill in when declaring. For example, for log commodities, it is necessary to fill in information such as "Type (Chinese, Latin species / genus name)", "Processing method", and "Minimum cross-sectional size", while for plate commodities, it is necessary to fill in information such as "Type (Chinese, Latin species / genus name)", "Specification", "Thickness", and "Processing method". The information filled in for commodities with different tax numbers is slightly different. In one embodiment, in order to unify the model input format, the acquired historical declaration data of customs commodities are subjected to field parsing to extract the field information common to different commodities, for example, the "Specification Elements" field is parsed. In one embodiment, the parsing process can, for example, match the field items corresponding to different commodities through the "Customs Import and Export Commodity Standard Declaration Catalog and Interpretation" and extract the required fields by string segmentation.

[0030] In one embodiment, preprocessing the acquired dataset includes constructing a priori knowledge base of customs commodities. In one embodiment, the priori knowledge base of customs commodities may include a statistical database of various factors underlying the tax code classification standards for various commodities. Commodity tax code classification is based on specific criteria. For example, commodities may be classified based on the following: type, appearance, processing method, and intended use. For example, for timber commodities, the type of log used directly determines the commodity's tax code category. Based on timber knowledge and existing data, there is a strong correlation between the Latin name of the log and the Chinese name of the log. The Latin name is generally composed of "genus name" + "specific epithet", and this information corresponds to the "species (Chinese, Latin species / genus name)" field in the "Specification Elements" field. At the same time, combined with the Latin-related data on timber in the dataset, a priori knowledge base can be constructed, for example, including 311 tree genera and 1572 tree species. This knowledge base can include the Chinese name, genus name, species name of the log, and the corresponding tariff category of the log (for example, "radiata pine log", "tropical log", "red wood", etc.). Table 1 shows an example of a portion of the priori knowledge base constructed for timber commodities.

[0031] Table 1 Prior knowledge base (partial)

[0032] In one embodiment, the subcategory of a declared customs commodity can be determined based on a text similarity algorithm. During the preprocessing phase, a priori knowledge base for customs commodities has been constructed. Using the commodity declaration data provided by enterprises and / or individuals during customs declarations, such as commodity names and specifications, the corresponding subcategory of the commodity can be determined using a text similarity algorithm. For example, based on the priori knowledge base constructed for timber commodities, the corresponding log category of a declared timber commodity can be determined using the data in the customs declaration form provided by enterprises and / or individuals during customs declarations.

[0033] In one embodiment, the declaration data can be matched with the prior knowledge data in the prior knowledge base by using a text similarity algorithm based on edit distance (Jaro-Wikler), thereby reducing matching errors and improving matching efficiency, for example, especially when there are omissions or errors in the filled-in declaration data.

[0034] Edit distance, also known as Levenshtein distance, refers to the minimum number of edit operations required to convert two strings from one to the other. The greater the distance between them, the more different they are. The text similarity (SIM) algorithm based on edit distance is shown in Equation 1:

[0035] In one embodiment, in order to further reduce the matching error and improve the matching efficiency, a text similarity algorithm based on optimized edit distance is provided, wherein the text similarity (sim j- w ) The algorithm formula is shown in Equation 2: j-w =sim+l*p(1-sim) (Equation 2)

[0036] In Equations 1 and 2, |s1| and |s2| represent the lengths of strings s1 and s2, respectively, m represents the number of characters matched between the two strings, t represents half the number of transpositions, l represents the number of common prefix characters, and p represents a scaling factor constant.

[0037] Fig. 2 shows a flow chart of a method 200 for determining subcategories of customs commodities for tax number classification according to an embodiment of the present disclosure. As shown in Fig. 2 , the method 200 includes steps S210-S230.

[0038] S210: performing matching similarity calculation on the field information (e.g., commodity name and / or specification element information) extracted from the declaration data of the declared customs commodity and the prior knowledge data in the prior knowledge base;

[0039] S220: Compare the result of the matching similarity calculation with a predetermined matching similarity threshold;

[0040] S230: Determine, based on the comparison result, a customs commodity subcategory in the prior knowledge base that has the greatest matching similarity with the extracted field information.

[0041] In one example, we discuss timber commodities. It should be understood that the techniques disclosed herein can be applied to any other applicable customs commodities besides timber commodities. Figure 3 shows a flow chart of a method 300 for determining the wood species of timber commodities according to an embodiment of the present disclosure. Method 300 may include steps S310-S380. It should be understood that the method may include more or fewer steps, and this is not a limitation herein.

[0042] S310: Obtain information such as "category (Chinese, Latin species / genus name)" corresponding to the product in the product name goods_d and the parsed specification element cpc_c;

[0043] S320: Determine whether the Latin genus name is in the prior knowledge base;

[0044] S330: If it is determined that the Latin genus name is in the prior knowledge base, select the same genus name data from the prior knowledge base to perform matching similarity calculation;

[0045] S340: If it is determined that the Latin genus name is not in the prior knowledge base, then use the full amount of data in the prior knowledge base to perform matching similarity calculation;

[0046] S350: Calculate the matching similarity of the Chinese names of the logs corresponding to the products;

[0047] S360: If the Chinese matching similarity value S1 is not lower than the first matching similarity threshold T1, then the log species C1 in the prior knowledge base corresponding to the maximum Chinese matching similarity value (TOP1 value) is returned. In one example, the first matching similarity threshold T1 may be 0.8.

[0048] S370: If the Chinese matching similarity value S1 is lower than the first matching similarity threshold T1, for example, less than 0.8, then calculate the log Latin matching similarity S2;

[0049] S380: If the Latin matching similarity value S2 is higher than the above-mentioned log Chinese matching similarity value S1 by a first margin value V1, for example, S2 is 5 percentage points higher than S1, then the log species C2 in the prior knowledge base corresponding to the maximum Latin matching similarity value (TOP1 value) is returned; otherwise, the log species C1 in the prior knowledge base corresponding to the Chinese similarity TOP1 value is still returned.

[0050] Referring back to FIG. 1 , after the subcategories of the customs commodities for tax number classification are determined, at step S130 , the predicted tax numbers of the declared customs commodities are determined based on the determined subcategories.

[0051] In one embodiment, a subcategory of a customs commodity corresponds to a tax number, and the tax number of the commodity can be directly determined based on the subcategory determined for the declared commodity.

[0052] In another embodiment, a subcategory of customs goods corresponds to more than one tariff number, because in the tariff, the tariff number classification standard may be further based on various other specification elements of the customs goods, such as appearance shape, processing method, purpose, etc.

[0053] According to the classification characteristics of the commodity tax number in the tariff, that is, the first four digits of the tax number are used to determine the item of the commodity. Under the item, the subsequent tax number (for example, the last six digits of the tax number) is further determined according to one or more specification element classification standards. In the embodiment of the present disclosure, a first four digit tax number classification model and one or more specification element classification models can be constructed. The first four digits of the tax number classification model is used to determine the first four digits of the declared commodity based on the declared data. In one example, the first four digits of the declared commodity can be extracted from the declared tax number in the declared data. The one or more specification element classification models are used to determine the specification element classification results for tax number division. In one example, the above model can be constructed for timber products, where the first four digits of the tax code classification model can be used to determine whether the first four digits of the declared product's tax code are 4403. If so, the subsequent tax code is determined by the specification element model's determination of the log's cross-sectional dimensions, for example, whether the log's cross-sectional dimensions are greater than 15 cm. Furthermore, the first four digits of the tax code classification model can also be used to determine whether the first four digits of the declared product's tax code are 4407. If so, the subsequent tax code can be determined by the specification element model's determination of the end joining method of the board. In another example, the above model can be constructed for cotton products, where the first four digits of the tax code classification model can be used to determine whether the first four digits of a non-retail pure cotton product's tax code are 5205. The subsequent tax code can be determined by the specification element model's determination of the count thickness and whether it is single yarn or multi-ply yarn.

[0054] FIG4 shows a flow chart of a method 400 for determining the tax number of a declared commodity according to an embodiment of the present disclosure. As shown in FIG4 , the method 400 includes steps S410 to S430. It should be understood that the method may include more or fewer steps, and this document does not limit this.

[0055] S410: Using the classification model of the first four digits of the commodity tax number, determine the first four digits of the tax number of the declared commodity based on the declared data.

[0056] S420: Based on the determined first four digits of the tax number, one or more specification element classification models are used to determine the specification element classification results for tax number classification of the declared goods based on the declared data.

[0057] S430: Determine the predicted tax number of the declared goods based on the subcategories and specification element classification results determined for the declared goods.

[0058] In an embodiment of the present disclosure, a tax number correspondence table for one or more categories of customs commodities can be constructed based on the tax number classification standards for one or more categories of customs commodities. The tax number correspondence table is used to indicate the specific tax numbers of customs commodities of each subcategory. After determining the subcategory of the declared customs commodity, the tax number correspondence table can be used to determine the predicted tax number of the declared commodity based on the specification element information included in the declaration data of the declared commodity. Table 2 shows an example of a portion of the tax number correspondence table constructed for timber commodities. As shown in Table 2, the ten-digit tax number correspondence table for timber commodities is constructed based on three dimensions: the type of log corresponding to the timber commodity, the cross-sectional size classification result, and the end-combination discrimination result.

[0059] Table 2 Tax code correspondence table (partial)

[0060] In one embodiment, the declared commodity is a timber commodity, and one or more specification element classification models for timber commodities may include: a cross-sectional dimension classification model for classifying the cross-sectional dimensions of logs, and an end joint discrimination model for judging whether the ends of the boards are jointed.

[0061] In one embodiment, the first four digits of the tax number classification model can be constructed based on the BERT pre-training model. For example, by fine-tuning on its own data set, the specific construction process may include: generating training sample data; extracting the fields after data parsing in preprocessing, and splicing the fields extracted from the declared product name and rule elements; generating labels corresponding to the training data, and truncating the first four digits of the declared product tax number; building a classification model, using bert-base as pre-training, setting model parameters, and fine-tuning.

[0062] In one embodiment, the cross-sectional size classification model is trained for commodities with tax numbers beginning with 4403. The reporting method for this information in the declared data is not uniform, and the filling styles are as follows: greater than 15 cm, 15CM+, 15 cm+, 16-28CM, more than 15 cm, 160MM, 12-20 inches, more than 24 inches, 0.2M+, etc. This model needs to determine whether the minimum cross-sectional size of the corresponding log of the wood commodity exceeds 15CM based on the field information obtained from the declared data. In one embodiment, the cross-sectional size classification model is constructed based on a BERT pre-trained model. For example, a BERT-based binary classification model is constructed. The specific construction process may include: generating training sample data, extracting the "minimum cross-sectional size" field after the data is parsed in the pre-processing; generating labels corresponding to the training data, and manually labeling the extracted data; constructing a classification model, using bert-base as pre-training, setting model parameters, and fine-tuning.

[0063] In one embodiment, the end-join discrimination model is trained on commodities with tax ID numbers beginning with 4407. This model determines whether the commodity names and specification elements in the declared data contain "end" and "non-end." In one embodiment, the end-join discrimination model uses regular expressions to determine end-joins.

[0064] The above examples illustrate how to leverage a priori knowledge base built for timber commodities to identify the specific species of logs corresponding to these commodities using a text similarity algorithm. Simultaneously, a deep learning-based model was trained to classify the first four digits of the timber tax code, a cross-sectional dimension classification model, and an end-joint discrimination model. To determine the tax code for timber commodities, this disclosure uses three dimensions: the corresponding log species, the cross-sectional dimension classification results, and the end-joint discrimination results, enabling accurate identification of the timber commodity's tax code.

[0065] In an embodiment of the present disclosure, after determining the predicted tax number for the declared commodity, method 100 may further include comparing the predicted tax number for the declared commodity with the declared tax number in the declaration data to determine whether there are any anomalies in the current declaration. Figure 5 shows a flow chart of method 500 for detecting anomalies in tax number declarations for timber commodities according to an embodiment of the present disclosure. Method 500 may include steps S510-S590.

[0066] S510: Obtain the commodity name goods_d, specification element cpc_c, and commodity tax code hscode_10 filled in by the enterprise and / or individual when declaring;

[0067] S520: Calculate matching similarity using the prior knowledge base built through pre-training and the edit distance text similarity algorithm;

[0068] S530: Identify the log type corresponding to the product using the product subcategory identification process shown in FIG3 ;

[0069] S540: Based on the first four digits tax code classification model proposed in the above embodiment, the first four digits tax code of the declared goods are predicted, and the first four digits tax code hscode_4 is returned;

[0070] S550: If the first four digits of the tax code hscode_4 == 4403, then enter the cross-section size classification model and return the classification result a1;

[0071] S560: If the first four digits of the tax code hscode_4 == 4407, then enter the end-combination discrimination model and return the discrimination result b1;

[0072] S570: Using the pre-built tax code correspondence table, based on the return result a1 of the log species and / or cross-sectional size classification model and / or the return result b1 of the end-joint discrimination model, determine the ten-digit tax code hscode_p in the tax code correspondence table, i.e., the predicted tax code of the declared commodity;

[0073] S580: Compare the predicted tax number hscode_p and the declared tax number hscode_10 to see if they are consistent;

[0074] S590: Output the comparison result. If they are consistent, return a normal result; otherwise, return an abnormal result.

[0075] This disclosure proposes a method for determining customs commodity tax numbers based on deep learning. By comparing the tax number determined for a declared commodity with the actual declared tax number of the declared commodity, it is possible to detect abnormal tax number declarations for customs commodities. The deep learning model according to the embodiments of this disclosure can automatically predict the tax number of declared customs commodities and automatically complete the detection of abnormal tax number declarations, thereby reducing the workload of customs departments, improving the efficiency and accuracy of supervision, and thus maintaining fair trade and the stability of tax collection.

[0076] 6 is a schematic block diagram of an apparatus 600 for determining a commodity tax number according to an embodiment of the present disclosure. As shown in FIG6 , the apparatus 600 includes a data acquisition module 610 , a subcategory determination module 620 , and a tax number determination module 630 .

[0077] The data acquisition module 610 is configured to acquire the declaration data of the declared commodity.

[0078] The subcategory determination module 620 is configured to determine a subcategory for tax code classification of the declared commodity based on the declared data.

[0079] The tax number determination module 630 is configured to determine a predicted tax number of the declared commodity based on the determined subcategory.

[0080] In an embodiment of the present disclosure, the apparatus 600 may further include a first determination submodule, a second determination submodule, and a third determination submodule. The first determination submodule is configured to determine the first four digits of the tax number of the declared commodity based on the declared data using a classification model for the first four digits of the commodity tax number; the second determination submodule is configured to determine, based on the declared data, a specification element classification result for the declared commodity used for tax number classification using one or more specification element classification models based on the determined first four digits of the tax number; and the third determination submodule is configured to determine a predicted tax number for the declared commodity based on the subcategory determined for the declared commodity and the specification element classification result.

[0081] In an embodiment of the present disclosure, the apparatus 600 may further include a comparison module configured to compare the predicted tax number of the declared commodity with the declared tax number in the declaration data to determine whether there is any abnormality in the current declaration.

[0082] In an embodiment of the present disclosure, the apparatus 600 may further include modules for executing steps of various other methods according to the above embodiments. The specific implementation details of the apparatus 600 are the same as those described above for the various methods and will not be repeated here.

[0083] This disclosure proposes a device for determining customs commodity tax numbers based on deep learning. By comparing the tax number determined for a declared commodity with the actual declared tax number of the declared commodity, this device can detect abnormal tax number declarations for customs commodities. The deep learning model according to the embodiments of this disclosure can automatically predict the tax number of declared customs commodities and automatically detect abnormal tax number declarations, thereby reducing the workload of customs departments, improving regulatory efficiency and accuracy, and thus maintaining fair trade and the stability of tax collection.

[0084] FIG7 shows a block diagram of an example computing device 700 in which an apparatus for detecting abnormal declarations of customs commodity tax numbers according to various embodiments of the present disclosure can be used. Specifically, the computing device 700 shown in FIG7 includes one or more processors (or processor cores) 710, one or more memory / storage devices 720, and one or more communication resources 730, wherein each of these processors, memory / storage devices, and communication resources can be communicatively coupled via a bus 740 or other interface circuits. For embodiments utilizing node virtualization (e.g., network function virtualization (NFV)), a hypervisor 702 can be executed to provide an execution environment for one or more network slices / sub-slices to utilize the hardware resources of the computing device 700.

[0085] The processor 710 may include, for example, a processor 712 and a processor 714. The processor 710 may be, for example, a central processing unit (CPU), a reduced instruction set computing (RISC) processor, a complex instruction set computing (CISC) processor, a graphics processing unit (GPU), a digital signal processor (DSP) such as a baseband processor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a radio frequency integrated circuit (RFIC), another processor (including those discussed herein), or any suitable combination thereof.

[0086] The memory / storage device 720 may include main memory, disk storage, or any suitable combination thereof. The memory / storage device 720 may include, but is not limited to, any type of volatile, non-volatile, or semi-volatile memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, solid-state memory, etc.

[0087] The communication resources 730 may include interconnect or network interface controllers, components, or other suitable devices to communicate with one or more peripheral devices 704 or one or more databases 506 or other network elements via the network 708. For example, the communication resources 730 may include wired communication components (e.g., for coupling via USB, Ethernet, etc.), cellular communication components, near field communication (NFC) components, (or Low energy) components, components, and other communication components.

[0088] The instructions 750 may include software, a program, an application, an applet, an application, or other executable code for causing at least one of the processors 710 to perform various processes. The instructions 750 may reside, in whole or in part, within at least one of the processor 710 (e.g., in a cache of the processor), the memory / storage device 720, or any suitable combination thereof. In addition, any portion of the instructions 750 may be transferred from any combination of the peripheral device 704 or the database 706 to the hardware resources of the computing device 700. Thus, the memory of the processor 710, the memory / storage device 720, the peripheral device 704, and the database 706 are examples of computer-readable and machine-readable media.

[0089] For example, the instructions 750 may include a computer-readable program that causes the processor 710 to execute a method for detecting abnormal declarations of commodity tax numbers according to an embodiment of the present disclosure. When the instructions 750 are executed by the processor 710, the method for detecting abnormal declarations of commodity tax numbers according to an embodiment of the present disclosure is implemented, such as, for example, methods 100, 200, 300, 400, and 500.

[0090] According to an embodiment of the present disclosure, an apparatus for detecting abnormal declarations of commodity tax numbers for customs purposes may include a processor 710 and a memory / storage device 720. When the processor 710 executes instructions stored in the memory / storage device 720, a method for detecting abnormal declarations of commodity tax numbers for customs purposes according to an embodiment of the present disclosure is implemented, for example, methods 100, 200, 300, 400, and 500.

[0091] It should be understood that the present disclosure is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present disclosure is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present disclosure.

[0092] The above description is merely a specific embodiment of the present disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the scope of protection of the present disclosure is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and such modifications or replacements should be included in the scope of protection of the present disclosure.

Claims

1. A method for determining a commodity tax number, comprising: Obtain the declaration data of the declared goods; Determine the sub-category for tax number classification of the declared commodity based on the declared data; as well as A predicted tax number of the declared commodity is determined based on the sub-category.

2. The method according to claim 1, further comprising: Determine the first four digits of the tax number of the declared commodity based on the declared data by using the first four digits classification model of the tax number of the commodity; According to the determined first four digits of the tax code, using one or more specification element classification models, based on the declared data, determine the specification element classification result for the tax code classification of the declared commodity; as well as The predicted tax number of the declared commodity is determined based on the subcategory determined for the declared commodity and the specification element classification result.

3. The method according to claim 2, wherein: The first four-digit tax number classification model is built based on the BERT pre-training model.

4. The method according to any one of claims 1 to 3, further comprising: Obtaining a data set related to one or more categories of customs commodities, the data set comprising historical declaration data of the one or more categories of customs commodities and / or prior data related to the one or more categories of customs commodities; Preprocessing the data set to construct a priori knowledge base for the one or more categories of customs commodities; as well as A deep learning model is used to determine the subcategory of the declared commodity based on the declared data using the prior knowledge base.

5. The method according to claim 4, wherein: The preprocessing of the data set to construct a priori knowledge base for the one or more categories of customs commodities includes: Perform field parsing on the acquired historical declaration data of customs commodities to extract field information common to different commodities.

6. The method according to claim 4, wherein: The using the deep learning model to determine the subcategory of the declared commodity based on the declared data using the prior knowledge base includes: Performing matching similarity calculation on the field information extracted from the declared data and the prior knowledge data in the prior knowledge base; comparing the result of the matching similarity calculation with a predetermined matching similarity threshold; and Based on the comparison result, a customs commodity subcategory in the prior knowledge base having the greatest matching similarity with the extracted field information is determined.

7. The method according to claim 6, wherein: The matching similarity calculation is based on a text similarity algorithm based on edit distance.

8. The method according to claim 1, further comprising: The predicted tax number of the declared commodity is compared with the declared tax number in the declared data to determine whether there is any abnormality in the current declaration.

9. The method according to any one of claims 4 to 8, wherein: The prior data related to the one or more categories of customs commodities include expert knowledge and / or rules related to the names and / or classifications of the corresponding categories of customs commodities.

10. The method according to any one of claims 4 to 8, wherein: The historical declaration data includes customs import and export declaration form data, and the declaration form data at least includes a commodity name field and a specification element field.

11. The method according to claim 2, wherein: The declared commodity is a timber-type declared commodity, and the subcategory used for tax number classification is the type of log corresponding to the timber-type declared commodity.

12. The method according to claim 11, wherein: The one or more specification element classification models include: a cross-sectional dimension classification model for classifying the cross-sectional dimensions of logs, and an end joint determination model for determining whether the ends of a board are jointed.

13. The method according to claim 11, wherein: The cross-sectional size classification model is built based on the BERT pre-training model.

14. The method according to claim 11, wherein: The end combination discrimination model uses regular expressions to judge the ends.

15. The method according to claim 11, further comprising: Extracting the Chinese name and Latin name of the log of the wood declared commodity from the declared data; Perform matching similarity calculation on the Chinese name of the log and the prior knowledge data in the prior knowledge base to obtain a Chinese name similarity value S1; Compare S1 with a first matching similarity threshold T1; When S1 is not less than T1, it is determined that the log category of the wood-type declared commodity is the log category C1 corresponding to the Chinese name of the log with the maximum matching similarity value in the prior knowledge base; When S1 is lower than T1, the Latin name of the log is matched with the prior knowledge data in the prior knowledge base to calculate the similarity to obtain the Latin name similarity value S2; Compare S2 with S1; When S2 is higher than S1 by a first margin value, it is determined that the log category of the declared timber commodity is the log species C2 corresponding to the Latin name of the log with the maximum matching similarity value in the prior knowledge base; otherwise, it is determined that the log category of the declared timber commodity is C1.

16. The method according to claim 15, further comprising: Extracting the Latin genus name of the wood-related declared commodity from the declared data; determining whether the prior knowledge base includes the Latin genus name; When it is determined that the prior knowledge base includes the Latin genus name, selecting the same genus name data in the prior knowledge base for matching similarity calculation; When it is determined that the Latin genus name is not included in the prior knowledge base, all data in the prior knowledge base are selected to perform matching similarity calculation.

17. A device for determining a commodity tax number, comprising: processor; as well as A memory having instructions stored therein, which, when executed by the processor, implement the method according to any one of claims 1 to 16.

18. A computer-readable storage medium having instructions stored thereon, which when executed, implement the method according to any one of claims 1 to 16.

19. An apparatus for determining a commodity tax number, comprising means for executing the method according to any one of claims 1 to 16.

Images