International logistics e-commerce platform context translation detection method

Abstract

The invention discloses a context translation detection method for an international logistics e-commerce platform, and relates to the technical field of machine translation detection, and the method comprises the steps: constructing a dynamic knowledge base of the international logistics e-commerce platform, the knowledge base comprising a structured term sub-base and a context rule sub-base; the method comprises the following steps: receiving a to-be-detected source text and an initial translation text generated by a machine translation system, and identifying a current service node to which the source text belongs; executing a context translation detection algorithm driven by a dynamic knowledge base on the received text data; and generating a final detection report according to an algorithm output result, wherein the detection report annotates parts suggested to be corrected and translated wrongly in the initial translation text. The term deviation in machine translation can be automatically recognized and corrected, the manual proofreading cost is effectively reduced, and transaction disputes and logistics delay caused by translation errors are reduced.

Description

Technical Field

[0001] This invention relates to the field of machine translation detection technology, and in particular to a method for detecting contextual translation on an international logistics e-commerce platform. Background Technology

[0002] With the continued rapid growth of global e-commerce, international logistics e-commerce platforms have become the core digital infrastructure connecting global suppliers and buyers and reconstructing the global supply chain. On such platforms, efficient and accurate multilingual information exchange is key to ensuring the smooth operation of international trade. Product descriptions, technical parameters, logistics documents, contract terms, and instant messaging content all need to be accurately translated between users who speak different languages. Any translation deviation may lead to misunderstandings, transaction disputes, compliance risks, or even logistics delays, directly impacting core business efficiency and customer trust.

[0003] To address the challenges of multilingualism, existing technologies and solutions primarily focus on machine translation (MT), optical character recognition (OCR), and generative artificial intelligence (AIGC). Despite significant progress in these technologies, issues remain regarding the accuracy of translation in highly specialized and dynamically changing contexts within international logistics and e-commerce. International logistics e-commerce interactions involve a large number of technical terms, industry abbreviations, and localized expressions that change over time, across regions, and markets. For example, there's the "MSDS" (Material Safety Data Sheet) for chemicals, "temperature-controlled transport" for special logistics, and different names for the same product in different regions (e.g., "elevator" in North America, but "Aufzug" in Germany). General translation models lack targeted domain knowledge graphs and real-time updated terminology databases, making them highly susceptible to mistranslation, semantic incompleteness, or cultural mismatch. More seriously, dynamic information such as platform operating rules, promotional policies, and newly opened logistics routes can generate new terms and expressions that existing static translation systems cannot adapt to in a timely manner, leading to lags and discrepancies in information synchronization between the platform and users.

[0004] In summary, existing translation technologies used in international logistics e-commerce platforms suffer from a lack of in-depth domain knowledge, making it difficult to ensure the accuracy of key business information. This not only increases the cost of manual proofreading and error correction but also becomes a major obstacle hindering automated trade processes and impacting user experience. Therefore, there is an urgent need in this field for a translation detection method that can deeply integrate knowledge from the logistics e-commerce domain. Summary of the Invention

[0005] This invention provides a method for detecting contextual translation on international logistics e-commerce platforms, including: Step 1: Construct a dynamic knowledge base for the international logistics e-commerce platform. This knowledge base includes a structured terminology sub-base and a contextual rules sub-base. Step 2: Receive the source text to be detected and its initial translated text generated by the machine translation system, and identify the current business node to which the source text belongs; Step 3: Execute the dynamic knowledge base-driven context translation detection algorithm on the received text data: Based on the current business node, activate the corresponding subset of term pairs containing context rules from the dynamic knowledge base; traverse the source text and the initial translation text, identify the term pairs, and compare them with the activated subset of term pairs; Step 4: Generate the final detection report based on the comparison results. The detection report includes annotations in the initial translated text indicating suggested corrections and translation errors.

[0006] The aforementioned method for detecting contextual translation on international logistics e-commerce platforms involves the following sub-steps in its dynamic knowledge base data processing: Extract multilingual term pairs from platform historical orders, logistics documents and product lists, and associate each term pair with the source business node and matching degree weight. After the association is completed, store them in the structured term sub-library. Build contextual constraint rules for term pairs for each business node, and store them in the contextual rule sub-library after construction; Extract user-submitted correction requests for term pairs from user interaction logs, and dynamically update the structured terminology sub-library based on these correction request records.

[0007] The aforementioned method for detecting contextual translation on international logistics e-commerce platforms includes the following specific data processing flow for the contextual translation detection algorithm: The structured terminology sub-base selects all terminology pairs that originate from the current business node and have consistent language pair identifiers. Then, the contextual constraint rules associated with each terminology pair are selected from the contextual rule sub-base, forming an activated subset of terminology pairs containing contextual constraint rules. The source text and the initial translation text are segmented and aligned to form candidate term pairs, and a context lexical unit is constructed for each candidate term pair. For each candidate term pair, calculate its comprehensive similarity score with each reference term pair in the activated term subset, select the reference term pair with the highest comprehensive similarity score to bind with the candidate term pair, and output it as the comparison result.

[0008] The present invention also provides a context translation detection system for international logistics e-commerce platforms, comprising: a dynamic knowledge base construction module, a detection text receiving module, a context translation comparison module, and a detection report generation module; The dynamic knowledge base construction module is used to build and update the dynamic knowledge base of the international logistics e-commerce platform. This knowledge base includes a structured terminology sub-base and a contextual rules sub-base. The detection text receiving module is used to receive the source text to be detected and its initial translated text generated by the machine translation system, and to identify the current business node to which the source text belongs; The context translation comparison module is used to execute a dynamic knowledge base-driven context translation detection algorithm on the received text data: based on the current business node, it activates the corresponding subset of term pairs containing context rules from the dynamic knowledge base; it traverses the source text and the initial translation text, identifies the term pairs, and compares them with the activated subset of term pairs; The detection report generation module is used to generate a final detection report based on the comparison results. The detection report includes annotations in the initial translated text indicating suggested corrections and translation errors.

[0009] The beneficial effects achieved by this invention are as follows: by constructing a dynamically updated knowledge base in the logistics and e-commerce field, the accuracy and contextual adaptability of professional terminology translation are significantly improved; it can automatically identify and correct terminology deviations in machine translation, effectively reduce manual proofreading costs, reduce transaction disputes and logistics delays caused by translation errors, thereby enhancing the reliability of multilingual interaction on the platform and improving international trade efficiency and user experience. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0011] Figure 1 This is a flowchart of a context translation detection method for an international logistics e-commerce platform provided in Embodiment 1 of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Example 1 like Figure 1 As shown, Embodiment 1 of this application provides a method for detecting contextual translation on an international logistics e-commerce platform, including: Step S10: Construct a dynamic knowledge base for the international logistics e-commerce platform, which includes a structured terminology sub-base and a contextual rules sub-base; When initially constructing the dynamic knowledge base, it is necessary to retrieve multi-source heterogeneous data from the international logistics e-commerce platform's backend system for the past three years. After processing, the data is stored in the corresponding sub-database. Subsequent dynamic knowledge bases will periodically (in this embodiment, this is set to be collected every 24 hours) retrieve, process, and store newly generated multi-source heterogeneous data from the international logistics e-commerce platform's backend system to ensure the timeliness of the knowledge base. This dynamic knowledge base is the foundation for driving the context translation detection algorithm, and its data processing process is specifically divided into the following sub-steps: Step S11: Extract multilingual term pairs from platform historical orders, logistics documents and product lists, and associate each term pair with the source business node and matching degree weight. After the association is completed, store it in the structured term sub-library. Collect the following structured or semi-structured multilingual text data from the backend database of international logistics e-commerce platforms: historical order data: including fields such as product name, specifications, logistics method, and customs declaration description; logistics documents: including structured fields in documents such as waybills, bills of lading, customs declarations, and insurance policies; product lists: including fields such as product categories, attribute tags, and product descriptions. The above data is cleaned to remove irrelevant symbols, duplicate content, and privacy information, and then processed by word segmentation, part-of-speech tagging, and term alignment. The term alignment process refers to using a BERT-based multilingual sequence alignment model to identify the correspondence of the same entity (lexical) in different languages, forming term pairs of [source language term, target language term]. For professional abbreviations (such as MSDS), compound words (such as temperature-controlled transportation), and regional synonyms (such as Elevator / Aufzug), a combination of rule matching and statistical co-occurrence analysis is used to determine the alignment relationship. Of course, term pairs that already exist in the dynamic knowledge base are not stored and are simply removed as duplicate content.

[0014] Then, the following structured metadata is associated with each term pair: Source business node: This identifies which business node the term pair originates from. These business nodes are divided according to the platform's business processes, and a platform module-business node mapping table is constructed based on the division results for querying. In this embodiment, the nodes are divided into five categories: product publishing, order fulfillment, logistics tracking, customs declaration, and after-sales customer service; Language pair identifier: This identifies the source and target language types; Matching degree weight: This weight is reassigned each time data is retrieved, and its calculation formula is as follows:

[0015] in It is the matching weight of the i-th term pair. It is the initial matching weight of the i-th term pair, derived from the confidence score output by the aforementioned alignment model; It is the decay coefficient of the initial matching degree weight, which decreases as the frequency of term usage increases; These are the time-series decay factors. Multi-scenario consistency factor External knowledge verification factor The fusion weights; , The decay rate is a preset value, where t is the current timestamp. The timestamp of the last time the i-th term pair was judged to be a correct translation; , This indicates that within the current time window T (nearly 24 hours), the i-th term pair The number of times the translation is consistent with the machine translation result (considered as error-free). This indicates that within the current time window T, the i-th term pair The number of times the result is inconsistent with the machine translation (considered as a suggested correction or translation error). This represents the total number of times the i-th term appears within the time window T. To prevent small constants with a denominator of 0; This indicates that an external industry knowledge base (such as a logistics terminology database or a customs terminology database) provides the i-th term pair. The verification result, i.e., the target language term generated by the external industry knowledge base for the i-th source language term, and the term pair Similarity between target language terms.

[0016] Step S12: Build contextual constraint rules for term pairs for each business node, and store them in the contextual rule sub-library after construction; Based on the historical corpus and manually annotated data of each business node, the contextual constraint rules for each term are automatically generated using decision trees or rule templates. Specifically, they can be divided into three categories: grammatical rules (e.g., in the customs declaration node, the quantity unit of goods must use the international standard unit abbreviation), semantic rules (e.g., in the logistics tracking node, delivered must correspond to delivered and cannot be translated as delivered), and contextual rules (e.g., if hazardous appears in the context, package should be translated as dangerous goods package instead of ordinary package).

[0017] The generated contextual constraint rules are associated with the corresponding terms.

[0018] Step S13: Extract the user's correction request record for term pairs from the user interaction log, and dynamically update the structured term sub-library based on the correction request record; For a given term, if a user submits a correction request to the dynamic knowledge base within a time window T (the last 24 hours), that is, requests to modify the original term... Revised to new terminology First, the request content needs to be recorded, and then manually reviewed for adoption. The new terminology used will affect... It will not directly replace the original terminology pair Instead, it is stored as a new term pair, and its initial matching weight is adjusted by modifying the request record on the original term pair. The matching degree weights are obtained by a second weighting, and the second weighting formula is expressed as:

[0019] in For terminology The original matching weight, To The value after performing the double weighting, The preset global learning rate, For semantic correction distance factor, , For the original terminology pair Target language terms in For the revised terminology Target language terms in To return and The semantic similarity function (i.e., the cosine similarity between the two terms in the semantic vector space) is defined as follows: U represents the set of users who submitted correction requests for the term pair, and u is any user in the user set U. For user u, the historical revised adoption rate, For the aging decay rate, This is the current timestamp. For users The time for submitting the revision, It is a small constant that avoids a denominator of zero. For cross-user consistency factors, , This indicates that within a time window T, for the term... The number of users who proposed the same correction. This indicates that within the time window T, for the term... The number of users who proposed any corrective actions. For the preset consistency threshold, when hour Return 1.5, when hour Return 1 when hour Returns 0.5.

[0020] The above-mentioned double-weighted formula comprehensively evaluates the semantic value, user accuracy, timeliness, and group consensus of a single correction action, and achieves refined and adaptive updates of the matching degree weight.

[0021] It should be noted that the above update operation is a soft update mode for the dynamic knowledge base, used to adapt to users' personalized corrections and ensure the integrity of the knowledge base, because some platforms prefer to use the old translations out of habit. If switching to hard update mode, the latest adopted terminology will be used directly. Replace the original terminology pair And the new terminology direct inheritance Matching weight.

[0022] Step S20: Receive the source text to be detected and its initial translated text generated by the machine translation system, and identify the current business node to which the source text belongs; Through a standardized API interface, we receive JSON-formatted translation data packets transmitted from international logistics e-commerce platforms. The data packets contain source language text, machine-generated initial translation text, language pair identifiers, and three essential fields, where the language pair identifiers are used to mark the source language type and the target language type. After receiving the data, it is necessary to identify which business node the received data packet belongs to based on the URL path of the calling interface, and bind the identified node identifier to the received data packet; for example, if the URL path is "XX / product / release", it means that the interface belongs to the product module. Then, by querying the platform module-business node mapping table built when dividing the business nodes, it is possible to identify which business node the data packet belongs to.

[0023] Step S30: Execute the dynamic knowledge base-driven contextual translation detection algorithm on the received text data: Based on the current business node, activate the corresponding subset of term pairs containing contextual rules from the dynamic knowledge base; traverse the source text and initial translation text, identify the term pairs, and compare them with the activated subset of term pairs. Specifically, this involves the following sub-steps: Step S31: Filter out all term pairs from the current business node that have consistent language pair identifiers from the structured term sub-library, and then filter out the context constraint rules associated with each term pair from the context rule sub-library to form an activated subset of term pairs containing context constraint rules. Step S32: Perform word segmentation and term alignment processing on the source text and the initial translation text to form candidate term pairs, and construct a context lexical unit for each candidate term pair; Irrelevant symbols, duplicate content, and privacy information are removed from the source text and initial translated text. Then, word segmentation, part-of-speech tagging, and term alignment are performed to form candidate term pairs. Next, focusing on the source language term and target language term in each candidate term pair, M words (M=5 in this embodiment) are extracted from both the source and initial translated texts to form the context words for that candidate term pair, represented as follows: , These are context words extracted from the source text. These are the context words extracted from the initial translated text.

[0024] Step S33: For each candidate term pair, calculate its comprehensive similarity score with each reference term pair in the activated term subset, select the reference term pair with the highest comprehensive similarity score to bind with the candidate term pair, and output it as the comparison result; For each candidate term pair, first select term pairs with consistent source language terms from the activated term subset as reference term pairs. Then, calculate the comprehensive similarity score between the candidate term pair and each selected reference term pair. The calculation formula is as follows:

[0025] in For candidate term pairs and its j-th reference term pair The overall similarity score, For candidate term pairs Target language terms in For reference terminology Target language terms in This is a semantic similarity function that returns... and Cosine similarity in semantic vector space; This is a context rule validation function; if the candidate terms match... and its context vector Satisfying the reference terminology Relevant contextual constraint rules If the condition is met, return 1; otherwise, return -1. As a similarity enhancement factor, , For reference terminology Matching weight, and These are the mean and standard deviation of the matching weights for all terms in the activated subset. To prevent small constants with a denominator of zero, For recent conflict counts, record reference terms. The number of times the user corrected the data within the time window T (approximately 24 hours). Adjustable parameters (as set in this embodiment) 0.3 (0.1) It is the hyperbolic tangent function.

[0026] After the calculation is completed, the reference term pair with the highest comprehensive similarity score is selected, and it is bound to the candidate term pair. Its specific score is marked, and it is output as the comparison result of the candidate term pair.

[0027] It should be noted that the context rule verification function... It is coded by engineers according to preset rules, and different types of contextual constraint rules will call different coding programs for rule verification.

[0028] Step S40: Generate a final detection report based on the comparison results. The detection report includes annotations in the initial translated text indicating suggested corrections and translation errors. For the comparison results of each candidate term pair, the following decision process is performed: If the combined similarity score between the candidate term pair and the reference term pair is greater than the first threshold, the translation is deemed error-free and ignored. If the combined similarity score between the candidate term pair and the reference term pair is less than the first threshold but greater than the second threshold, it is determined to be a correction suggestion. In the initial translation text, the target language term is replaced with the target language term in the reference term pair, and the replaced part is highlighted and prompted with "System suggests correction". If the combined similarity score between the candidate term pair and the reference term pair is less than the second threshold, it is judged as a translation error and marked in more prominent red in the initial translated text.

[0029] After making independent decisions on all candidate term pairs, the entire annotated text is scanned to check for inconsistencies in the translation of the same source term into different target terms in the context. If such inconsistencies are found, the translation with the highest weight is unified through a voting mechanism based on the matching weight of these target terms in the knowledge base and the current comprehensive similarity score, thus ensuring the consistency of terminology within the document.

[0030] The first and second thresholds are calibrated based on the actual test results.

[0031] Example 2 Embodiment 2 of this application provides a context translation detection system for an international logistics e-commerce platform, including: a dynamic knowledge base construction module, a detection text receiving module, a context translation comparison module, and a detection report generation module; (1) Dynamic knowledge base construction module, used to build and update the dynamic knowledge base of the international logistics e-commerce platform. The knowledge base includes a structured terminology sub-base and a contextual rules sub-base. (2) A detection text receiving module, used to receive the source text to be detected and its initial translated text generated by the machine translation system, and to identify the current business node to which the source text belongs; (3) Context translation comparison module, used to execute dynamic knowledge base driven context translation detection algorithm on the received text data: based on the current business node, activate the corresponding term pair subset containing context rules from the dynamic knowledge base; traverse the source text and the initial translation text, identify the term pairs therein, and compare them with the activated term pair subset; (4) The test report generation module is used to generate the final test report based on the comparison results. The test report includes annotations in the initial translated text that suggest corrections and identify translation errors.

[0032] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor; The memory is used to store one or more program instructions; A processor is used to run one or more program instructions to execute a context translation detection method for an international logistics e-commerce platform.

[0033] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide a method for detecting contextual translation on an international logistics e-commerce platform.

[0034] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the aforementioned method for detecting contextual translation on an international logistics e-commerce platform.

[0035] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0036] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0037] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0038] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0039] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0040] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0041] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0042] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting contextual translation on an international logistics e-commerce platform, characterized in that, include: Step 1: Construct a dynamic knowledge base for the international logistics e-commerce platform. This knowledge base includes a structured terminology sub-base and a contextual rules sub-base. Step 2: Receive the source text to be detected and its initial translated text generated by the machine translation system, and identify the current business node to which the source text belongs; Step 3: Execute the dynamic knowledge base-driven context translation detection algorithm on the received text data: Based on the current business node, activate the corresponding subset of term pairs containing context rules from the dynamic knowledge base; traverse the source text and the initial translation text, identify the term pairs, and compare them with the activated subset of term pairs; Step 4: Generate the final detection report based on the comparison results. The detection report includes annotations in the initial translated text indicating suggested corrections and translation errors.

2. The method for detecting contextual translation on international logistics e-commerce platforms according to claim 1, characterized in that, When building a dynamic knowledge base for the first time, it is necessary to pull multi-source heterogeneous data from the back-end system of the international logistics e-commerce platform for the past three years, process it, and store it in the corresponding sub-database. Subsequent dynamic knowledge bases will periodically pull, process, and store newly generated multi-source heterogeneous data from the back-end system of the international logistics e-commerce platform.

3. The method for detecting contextual translation on international logistics e-commerce platforms according to claim 2, characterized in that, The specific data processing process of a dynamic knowledge base can be divided into the following sub-steps: Extract multilingual term pairs from platform historical orders, logistics documents and product lists, and associate each term pair with the source business node and matching degree weight. After the association is completed, store them in the structured term sub-library. Build contextual constraint rules for term pairs for each business node, and store them in the contextual rule sub-library after construction; Extract user-submitted correction requests for term pairs from user interaction logs, and dynamically update the structured terminology sub-library based on these correction request records.

4. The method for detecting contextual translation on international logistics e-commerce platforms according to claim 1, characterized in that, The specific data processing flow of the context translation detection algorithm is as follows: The structured terminology sub-base selects all terminology pairs that originate from the current business node and have consistent language pair identifiers. Then, the contextual constraint rules associated with each terminology pair are selected from the contextual rule sub-base, forming an activated subset of terminology pairs containing contextual constraint rules. The source text and the initial translation text are segmented and aligned to form candidate term pairs, and a context lexical unit is constructed for each candidate term pair. For each candidate term pair, calculate its comprehensive similarity score with each reference term pair in the activated term subset, select the reference term pair with the highest comprehensive similarity score to bind with the candidate term pair, and output it as the comparison result.

5. The method for detecting contextual translation on international logistics e-commerce platforms according to claim 4, characterized in that, Contextual lexicons refer to data pairs formed by extracting M words from the source text and the initial translation text, respectively, centered on the source language term and the target language term in the candidate term pair.

6. The method for detecting contextual translation on international logistics e-commerce platforms according to claim 4, characterized in that, For each candidate term pair, its reference term pair is a term pair that matches the source language terms selected from the activated term subset.

7. A context translation detection system for international logistics e-commerce platforms, characterized in that, Includes: a dynamic knowledge base construction module, a text receiving module, a context translation comparison module, and a detection report generation module; The dynamic knowledge base construction module is used to build and update the dynamic knowledge base of the international logistics e-commerce platform. This knowledge base includes a structured terminology sub-base and a contextual rules sub-base. The detection text receiving module is used to receive the source text to be detected and its initial translated text generated by the machine translation system, and to identify the current business node to which the source text belongs; The context translation comparison module is used to execute a dynamic knowledge base-driven context translation detection algorithm on the received text data: based on the current business node, it activates the corresponding subset of term pairs containing context rules from the dynamic knowledge base; it traverses the source text and the initial translation text, identifies the term pairs, and compares them with the activated subset of term pairs; The detection report generation module is used to generate a final detection report based on the comparison results. The detection report includes annotations in the initial translated text indicating suggested corrections and translation errors.

Images