Order address recognition system based on semantic recognition technology

Abstract

The invention relates to the technical field of logistics transportation, in particular to an order address recognition system based on a semantic recognition technology, comprising a text preprocessing module which is used for segmenting text content through rules and statistical algorithms based on each segment of text content in an order address, screening out meaningless words, single characters and special symbols, and processing the text content in the order address; generating an order text preprocessing set; the language type recognition module is used for determining a corresponding language type through a language detection model based on the order text preprocessing set, translating the corresponding language type into a specified type of language, associating different language texts through mapping, sorting order addresses into a unified language, and generating a basic output text; according to the method, statistical analysis is carried out on the order addresses after optimization and calibration, the association degree of the addresses, the personnel and the commodities can be evaluated in combination with the time period, so that the orders which are maliciously hoarded through intentionally fuzzy receiving addresses are recognized, and the purpose of maintaining the market stability is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics transportation, and particularly to an order address recognition system based on semantic recognition technology. Background Art

[0002] An order address usually includes the consignee's name, contact phone number, detailed address, and remarks information, etc. In the logistics business, accurate order addresses can be used to combine multiple packages for delivery to the same target area, thereby optimizing the itinerary and time arrangement. Therefore, the order address is crucial in cross-regional delivery and is a key factor in reducing logistics delays and improving the customer experience.

[0003] However, in the actual logistics transportation process, problems such as unclear addresses or spelling mistakes may occur. At this time, it may lead to the package not being delivered on time or even being delivered to the wrong place. In addition, for some malicious hoarding behaviors, some people may deliberately provide vague address information to confuse the public, thereby reducing the probability of the packages being recognized as belonging to the same person. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art and to propose an order address recognition system based on semantic recognition technology.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: An order address recognition system based on semantic recognition technology, comprising:

[0006] A text preprocessing module: Based on each piece of text content in the order address, the text content is segmented through rules and statistical algorithms, meaningless words, single characters, and special symbols are screened out, the valid text content is retained and corrected, and an order text preprocessing set is generated;

[0007] A language type recognition module: Based on the order text preprocessing set, the corresponding language type is determined through a language detection model, and at the same time, corresponding parsing is performed, translated into a specified type of language, different language texts are associated through mapping, and the order address is sorted into a unified language to generate a basic output text;

[0008] A semantic recognition module: Based on the basic output text, the lexical relationships and long-term and short-term dependencies in the context are analyzed through a deep learning model, the semantics of the text content is understood, the integrity and validity of the address information are judged, and the basic administrative division units missing in the address composition are initially recognized to generate an order content recognition report;

[0009] An address sorting module: Based on the order content recognition report, the typos in the text content are corrected according to the built-in dictionary, the basic administrative division units missing in the order address are supplemented, and the address composition is adjusted in a set order to generate a standard order address set;

[0010] Address verification module: Based on the standard order address set, match the corresponding address space coordinates through a GIS map tool, compare the composition of the input address with the address composition provided by GIS, perform reverse geocoding, infer missing parts, modify conflicting parts, form a complete order address, and generate an optimized set of address information;

[0011] User communication module: Based on the optimized set of address information, prompt the abnormal composition range of the order address through a human-computer interaction interface, provide correction suggestions, allow users to correct the address and select recommended addresses, and perform secondary confirmation after modification to generate a final order address table;

[0012] Malicious order identification module: Based on the final order address table, count the personnel information of the order address, calculate the usage frequency corresponding to each order address at the same time, identify high-frequency and low-frequency addresses, retrieve the corresponding order details according to the high-frequency used addresses and personnel information, check the purchased goods of each batch of orders, and generate an order statistics report;

[0013] Risk prediction module: Based on the order statistics report, analyze the correlation degree among the order address, personnel information, and purchased goods through machine learning and data analysis techniques, screen out abnormal orders by combining anomaly detection, time series analysis, classification algorithms, and clustering algorithms, evaluate the probability of malicious hoarding, and generate a risk identification report;

[0014] Data processing module: Based on the risk identification report, collect system operation logs, perform format conversion and record backup on known information, and generate a system data processing report.

[0015] As a further solution of the present invention, the text preprocessing module includes a text division unit, a word deletion unit, and a specification correction unit, where:

[0016] Text division unit: Based on the input text content, perform multi-step and multi-dimensional segmentation on the text content through rules and statistical algorithms, divide the irregular order address into multiple basic texts, and generate a text division report;

[0017] Word deletion unit: Based on the text division report, clean the text content through regular expressions, stemming, and lemmatization to reduce redundancy and generate a text cleaning report;

[0018] Specification correction unit: Based on the text cleaning report, perform letter conversion and correction through a natural language processing model to maintain the unity of the text content and generate a preprocessed set of order texts.

[0019] As a further solution of the present invention, the language type recognition module includes a language detection unit, an analysis and fusion unit, and a language conversion and output unit, where:

[0020] The language detection unit; based on the preprocessed set of order texts, extracts text features through Langid, including character frequencies and letter sequence combinations, counts the probabilities of the occurrence of specific words or combinations of words in each language, generates a classification model for each language in cooperation with the Naive Bayes algorithm, screens out the corresponding text languages, determines the required translation models, and generates a text language screening report;

[0021] The analysis and fusion unit: based on the text language screening report, performs text translation through the corresponding translation model, uses a machine learning model in cooperation with a mapping table to match different text languages, and generates a text content association set;

[0022] The language conversion and output unit: based on the text content association set, compares and converts the input different text languages with the standard format, performs cross-language text integration, and generates a basic output text.

[0023] As a further solution of the present invention, the semantic recognition module includes a database construction unit, a model training unit, and a semantic understanding unit, where:

[0024] The database construction unit: based on the basic output text, maps the data to a high-dimensional space through a deep learning model, performs vector conversion of sentences and words on the stored standard order texts, and embeds them in the space after annotation waiting for subsequent matching, generating a basic data set;

[0025] The model training unit: based on the basic data set, in cooperation with experimental orders and historical actual orders with correct and incorrect address information, trains a learning model through a supervised learning algorithm, exercises the ability to predict missing content in the text, and adjusts the model parameters through cross-validation, accuracy evaluation, and loss function optimization, generating a model training library;

[0026] The semantic understanding unit: based on the model training library, matches highly similar entries, analyzes the input text content through natural language processing technology, extracts relevant information related to the address, judges the integrity and consistency of the information, and supplements it in combination with the context, generating an order content recognition report.

[0027] As a further solution of the present invention, the address sorting module includes a spelling correction unit, an element supplement unit, and a standardization and sorting unit, where:

[0028] Spelling correction unit: Based on the order content recognition report, match the vocabulary in the order content through the built-in dictionary, identify the words with inaccurate spelling, and according to the preset dictionary rules, replace the incorrect spelling with the correct spelling to generate an address correction report;

[0029] Element supplementation unit: Based on the address correction report, identify the missing administrative division units in the text content through named entity recognition technology, speculate and fill in the missing administrative division units according to the specific address name and context information to generate an address supplementation report;

[0030] Standardization and arrangement unit: Based on the address supplementation report, adjust the order of the address components according to the hierarchical relationship constraints, and unify the separators in the address into the standard form to generate a set of standard order addresses.

[0031] As a further solution of the present invention, the address verification module includes a geocoding unit, a matching and comparison unit, and a depth optimization unit, wherein:

[0032] Geocoding unit: Based on the set of standard order addresses, parse the input text content, extract the relevant address information of the street, city, and country, combine with the map data in the GIS database, and convert the given standard order address into the corresponding geospatial coordinates through the geocoding algorithm to generate a coordinate conversion report;

[0033] Matching and comparison unit: Based on the coordinate conversion report, match and compare the order address with the address provided by the GIS standard address library through the string matching algorithm and the geographical location similarity analysis, check the differences and conflicts between the two, identify the missing information, redundant parts, and spelling errors still existing in the order address, and mark them to generate an address matching report;

[0034] Depth optimization unit: Based on the address matching report, perform reverse coding according to the coordinate points through the machine learning algorithm, reverse infer and fill in the detailed text content of the order address, and correct the non-matching place names to generate an optimized set of address information.

[0035] As a further solution of the present invention, the user communication module includes an error prompt unit, a user feedback unit, and a modification confirmation unit, wherein:

[0036] Error prompt unit: Based on the optimized set of address information, highlight the abnormal areas and fields of the order address through the UI interface, provide a clear description of the correction range, inform the possibility and reason of the error, assist the user to identify the problem and make corresponding corrections, and generate a problem address prompt report;

[0037] User feedback unit: Based on the problem address hint report, provide correction suggestions provided by the system for the user to select through the UI interface. For the correction suggestions that are not completely accurate, the user manually modifies the address to generate an address feedback report.

[0038] Modification confirmation unit: Based on the address feedback report, perform secondary confirmation on the modified content, display the modified address and the original address, generate the final order address, which is used as the basis for transportation, delivery, and inspection, and generate a final order address table.

[0039] As a further solution of the present invention, the malicious order recognition module includes a personnel information recognition unit, an address frequency statistics unit, and an order details retrieval unit, where:

[0040] Personnel information recognition unit: Based on the final order address table, extract the personnel information related to the address in the order text content, remove duplicate and incomplete information through cleaning, and count the frequently used personnel information through a basic counting algorithm to generate a personal information collation report.

[0041] Address frequency statistics unit: Based on the final order address table, use a prefix tree and counting to statistically analyze the usage frequency of each order address, sort the addresses according to the time period and frequency, identify high-frequency addresses and low-frequency addresses, and mark the high-frequency addresses to generate an address frequency statistics table.

[0042] Order details retrieval unit: Based on the personal information collation report and the address frequency statistics table, analyze the correlation degree between high-frequency addresses and personnel information through a clustering algorithm, retrieve the orders corresponding to high-frequency addresses and the same personnel information, summarize the specific commodity information and order placement time of each order, and generate an order statistics report.

[0043] As a further solution of the present invention, the risk prediction module includes an information matching unit, a historical information screening unit, and a risk assessment unit, where:

[0044] Information matching unit: Based on the order statistics report, analyze the patterns between commodities, order addresses, and personnel information within the time period through cosine similarity, analyze the possibility of association, identify potential malicious behaviors, and generate a malicious order analysis report.

[0045] Historical information screening unit: Based on the malicious order analysis report, screen the historical order data of the corresponding account, analyze the time series of historical orders, identify the normal fluctuation patterns of purchase behaviors, and generate a historical order analysis report.

[0046] Risk assessment unit: Based on the historical order analysis report, perform anomaly detection on the order information within the current time period, identify the orders that deviate from normal behaviors through a classification algorithm, and generate a risk identification report.

[0047] As a further solution of the present invention, the data processing module includes a log and monitoring unit, a data conversion unit, and an upload and backup unit, where:

[0048] Log and monitoring unit: Based on the risk identification report, collect and monitor the operation logs, error logs, and security logs of the system, real-time monitor the system status, discover potential security problems and performance bottlenecks, and generate a system log report;

[0049] Data conversion unit: Based on the system log report, convert the formats of known data from different sources into a format that meets specific requirements, ensure data consistency, and generate a standardized data set;

[0050] Upload and backup unit: Based on the standardized data set, regularly back up to the cloud, and through encryption and version control, ensure the effectiveness and recoverability of the data, and generate a system data processing report.

[0051] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0052] 1. In the present invention, by statistically analyzing the optimized and calibrated order addresses, the usage frequencies of each precise address and the corresponding personnel information can be obtained. Furthermore, by combining the time period, the correlation degree among the address, personnel, and goods can be evaluated to identify orders that maliciously hoard goods by deliberately obscuring the receiving address, improving the detection accuracy, achieving the purpose of maintaining market stability, and ensuring the reasonable allocation of logistics resources.

[0053] 2. In the present invention, by performing language conversion processing on the order text before address verification, the address information in multiple languages can be uniformly converted into the target text language, thereby ensuring the smooth processing of international orders, effectively avoiding address recognition errors caused by language mixing, and adapting to the language requirements of different regions through language conversion to ensure the accuracy and consistency of address information during cross-border delivery.

[0054] 3. In the present invention, by deeply analyzing and identifying the text content of the order address through machine learning and comparing it with the standard address information in the database, the accuracy of the address is automatically verified, reducing the error rate during address entry. Furthermore, the situations of delivery failure or delay caused by address errors are reduced. At the same time, common spelling mistakes and format problems can be automatically identified and corrected, thereby optimizing the delivery process and improving the delivery efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is the system flow chart of the present invention;

[0056] Figure 2 is the schematic diagram of the system framework of the present invention;

[0057] Figure 3 Schematic diagram of the malicious order recognition module of the present invention. Specific embodiments

[0058] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] Embodiment 1

[0060] Please refer to Figure 1 , the present invention provides a technical solution: an order address recognition system based on semantic recognition technology, including:

[0061] Text preprocessing module: Based on each piece of text content in the order address, the text content is segmented through rules and statistical algorithms, meaningless words, single characters, and special symbols are screened out, the effective text content is retained and corrected, and an order text preprocessing set is generated. After preprocessing the text content of the order address according to the algorithm, the quality and structured degree of the text content can be significantly improved, so as to prepare for subsequent data analysis, logistics processing, etc.;

[0062] Language type recognition module: Based on the order text preprocessing set, the corresponding language type is determined through a language detection model, and at the same time, corresponding parsing is performed, translated into a specified type of language, different language texts are associated through mapping, the order address is sorted into a unified language, and a basic output text is generated. Through automatic language recognition, translation, mapping and sorting, the text content can be processed in a unified language environment, and then the delivery of international orders can be effectively arranged, the interference caused by language differences can be reduced, and the efficiency and accuracy of order processing can be improved;

[0063] Semantic recognition module: Based on the basic output text, the lexical relationships and long-term and short-term dependencies in the context are analyzed through a deep learning model, the semantics of the text content is understood, the integrity and validity of the address information are judged, and the basic administrative division units missing in the address composition are initially identified, and an order content recognition report is generated. Through deep learning technology, the semantics of the input text content are analyzed and judged, and then the integrity, validity and accuracy of the text content can be quickly recognized in cooperation with the trained model, improving the processing efficiency of the text content;

[0064] Address sorting module: Based on the order content recognition report, correct typos in the text content according to the built-in dictionary, supplement the missing administrative division basic units in the order address, adjust the address composition in the set order, and generate a standard order address set. Through automatic error correction, supplementation, and format adjustment, text content in different forms can be adjusted to conform to the standard processing format, thereby facilitating subsequent comparison with actual coordinates;

[0065] Address verification module: Based on the standard order address set, match the corresponding address space coordinates through the GIS map tool, compare the input address composition with the address composition provided by GIS, perform reverse geocoding, infer missing content, modify conflicting content, form a complete order address, and generate an address information optimization set. Based on the GIS map tool, the input order address can be compared with the address provided by GIS search, and the missing address part can be inferred automatically when the address is incomplete. At the same time, conflicting address content can also be corrected, so as to achieve the purpose of verifying the accuracy and integrity of the order address;

[0066] User communication module: Based on the address information optimization set, prompt the abnormal composition range of the order address through the man-machine interaction interface, provide correction suggestions, allow users to correct the address and select recommended addresses, and perform secondary confirmation after modification to generate the final order address table. When an address anomaly is identified, the problems found can be shown to the user through interaction with the user, and corresponding modifications can be made to ensure the accuracy of the final order address, prevent the existence of ambiguous order addresses from affecting actual delivery, and avoid ambiguous receiving addresses from affecting subsequent identification of malicious hoarding;

[0067] Malicious order identification module: Based on the final order address table, count the personnel information of the order address, calculate the usage frequency corresponding to each order address at the same time, identify high-frequency and low-frequency addresses, retrieve the corresponding order details according to the high-frequency used addresses and personnel information, check the purchased goods of each batch of orders, and generate an order statistical report. By statistically analyzing the personnel information and usage frequency of order addresses within a set period, abnormal order information can be screened out, thereby providing a basis for subsequent identification of potential malicious orders;

[0068] Risk prediction module: Based on the order statistical report, analyze the correlation degree among the order address, personnel information, and purchased goods through machine learning and data analysis techniques. Combine anomaly detection, time series analysis, classification algorithms, and clustering algorithms to screen out abnormal orders, evaluate the probability of malicious hoarding, generate a risk identification report, and analyze the relationship among the order address, personnel information, and purchased goods through machine learning and data analysis techniques, so as to evaluate the screened risk orders to confirm the orders with malicious hoarding behavior, and can cooperate with the logistics system to report and actually inspect such orders in the follow-up to maintain market stability and avoid the occupation of logistics resources;

[0069] Data processing module: Based on the risk identification report, collect the system operation logs, perform format conversion and record backup on the known information, and generate a system data processing report. By recording and uploading the internal data of the system, the reliability and integrity of the data can be ensured, and relevant data can be retrieved for maintenance investigation in case of problems, which guarantees the stable operation of the system and ensures the traceability of the data.

[0070] Please refer to Figure 2 , the text preprocessing module includes a text division unit, a word and character deletion unit, and a specification correction unit, where:

[0071] Text division unit: Based on the input text content, perform multi-step and multi-dimensional segmentation on the text content through rules and statistical algorithms, divide the irregular order address into multiple basic texts, and generate a text division report. By splitting the text into smaller units, it is convenient for subsequent entity recognition, translation, and hierarchical analysis of the text content, etc., and thus can lay a good foundation for subsequent efficient and accurate natural language processing tasks;

[0072] Word and character deletion unit: Based on the text division report, clean the text content through regular expressions, stemming, and lemmatization to reduce redundancy and generate a text cleaning report. Through relevant algorithms, useless or duplicate text data can be identified and deleted, reducing the redundant content of the text;

[0073] Specification correction unit: Based on the text cleaning report, perform letter conversion and correction through a natural language processing model to maintain the unity of the text content and generate an order text preprocessing set. By processing the letters of some orders containing foreign languages, the consistency of the text content can be ensured, preventing the subsequent text processing from being affected by capitalization, etc.

[0074] Please refer to Figure 2 , the language type recognition module includes a language detection unit, an analysis and fusion unit, and a language conversion and output unit, where:

[0075] Language detection unit: Based on the preprocessed order text set, extract text features through Langid, including character frequencies and letter sequence combinations, count the probabilities of specific words or word combinations in each language, generate a classification model for each language in combination with the Naive Bayes algorithm, screen out the corresponding text languages, determine the required translation models, generate a text language screening report. The Langid tool and the Naive Bayes algorithm can effectively extract the language features of the text, and then each language text can be screened and marked for subsequent translation and integration;

[0076] Parsing and integration unit: Based on the text language screening report, perform text translation through the corresponding translation model, use a machine learning model in combination with a mapping table to match different text languages, generate a text content association set, translate the text content in different languages into the same language through the matching translation model, and then the machine learning model and the mapping table can mark the text content in different languages with the same meaning for subsequent integration and sorting;

[0077] Language conversion and output unit: Based on the text content association set, compare and convert the input different text languages with the standard format, perform cross-language text integration, generate the basic output text. By jointly using pre-trained multi-language embedding models, cross-language pre-trained models, and multi-language generation models, etc., different text languages can be integrated and converted into the standard format, so as to prepare for the subsequent processing of international orders and avoid address recognition errors caused by language mixing.

[0078] Please refer to Figure 2 , the semantic recognition module includes a database construction unit, a model training unit, and a semantic understanding unit, where:

[0079] Database construction unit: Based on the basic output text, map the data to a high-dimensional space through a deep learning model, perform vector conversion of sentences and words on the stored standard order text, and embed it in the space after annotation waiting for subsequent matching, generate a basic data set. Through the constructed vectorized database, similarity search, nearest neighbor query, and large-scale vector data management, etc. can be performed to facilitate subsequent model training and data matching;

[0080] Model training unit: Based on the basic data set, in combination with experimental orders and historical actual orders with correct and incorrect address information, train the learning model through a supervised learning algorithm, exercise the ability to predict missing content in the text, and adjust the model parameters through cross-validation, accuracy evaluation, and loss function optimization, generate a model training library. Train the text prediction model through a large number of diversified and labeled order address training texts, so that the learning model can be fully applicable to the recognition and analysis of different text contents, and at the same time, the learning model can be continuously improved through relevant optimization methods;

[0081] Semantic Understanding Unit: Based on the model training library, it matches highly similar entries, analyzes the input text content through natural language processing technology, extracts relevant information related to the address, judges the integrity and consistency of the information, and makes supplements in combination with the context to generate an order content recognition report. It processes the sequence data through the RNN model and the BERT model to capture the dependencies of words in terms of time and order, and then cooperates with the LDA model to mine the potential topic distribution in the text to assist in understanding the core content of the text and prepare for subsequent corrections.

[0082] Please refer to Figure 2 , the address sorting module includes a spelling correction unit, an element supplementation unit, and a standardization sorting unit, where:

[0083] Spelling Correction Unit: Based on the order content recognition report, it matches the vocabulary in the order content through the built-in dictionary, identifies the misspelled vocabulary, and replaces the incorrect spelling with the correct spelling according to the preset dictionary rules to generate an address correction report. The use of the built-in dictionary can ensure the reliability of the data, making the text content easier to understand and use in subsequent processing;

[0084] Element Supplementation Unit: Based on the address correction report, it identifies the missing administrative division units in the text content through named entity recognition technology, speculates and fills in the missing administrative division units based on the specific name of the address and the context information to generate an address supplementation report. By supplementing the missing administrative division units, the text content can be made more complete, improving the accuracy of the order address for subsequent correct delivery;

[0085] Standardization Sorting Unit: Based on the address supplementation report, according to the hierarchical relationship constraints, it adjusts the address composition order according to the set administrative division tree and unifies the separators in the address into the standard form to generate a standard order address set. Through the set administrative division tree, the text content can be adjusted in the preset structural form, making it easier for subsequent GIS map tools, etc. to correctly understand the order address. At the same time, for some areas of the text, it can be separated by spaces or commas to ensure the clarity of the address.

[0086] Please refer to Figure 2 , the address verification module includes a geocoding unit, a matching and comparison unit, and a depth optimization unit, where:

[0087] Geocoding Unit: Based on the standard order address set, it parses the input text content, extracts relevant address information such as streets, cities, and countries, combines the map data in the GIS database, and converts the given standard order address into the corresponding geospatial coordinates through the geocoding algorithm to generate a coordinate conversion report. By converting the end point of the sorted order address into geographical coordinates, i.e., longitude and latitude, through the GIS database, the order address can be made concrete and standardized;

[0088] Matching and comparison unit: Based on the coordinate transformation report, through string matching algorithms and geographical location similarity analysis, it matches and compares the order address with the addresses provided in the GIS standard address library, checks for differences and conflicts between the two, identifies missing information, redundant parts, and spelling mistakes still existing in the order address, marks them, generates an address matching report. By comparing the longitude and latitude of the end point of the order address with the search results in the GIS standard address library, the correctness and accuracy of the order address can be judged. At the same time, by analyzing the degree of geographical location difference, it can also identify whether there are streets with the same name in different cities or cities with similar names based on geographical significance;

[0089] Deep optimization unit: Based on the address matching report, through machine learning algorithms, it performs reverse coding according to coordinate points, reverse infers and fills in the detailed text content of the order address, and corrects the unmatched place names, generating an optimized set of address information. Through reverse inference, it can accurately fill in the missing parts of the input address according to the search results in the GIS standard address library and correct the unmatched place names entered;

[0090] Please refer to Figure 2 , the user communication module includes an error prompt unit, a user feedback unit, and a modification confirmation unit, where:

[0091] Error prompt unit: Based on the optimized set of address information, it highlights the abnormal areas and fields of the order address through the UI interface, provides a clear description of the correction scope, informs the possibility and reason of the error, assists the user in identifying the problem and making corresponding corrections, generates a problem address prompt report. Through the UI interface, it can display the areas with input errors such as streets, towns, or cities, and display the address data retrieved by GIS, thus assisting the user in discovering possible incorrect order addresses;

[0092] User feedback unit: Based on the problem address prompt report, it allows the user to select the correction suggestions provided by the system through the UI interface. For the correction suggestions that are not completely accurate, the user manually modifies the address, generating an address feedback report. The recommended modified addresses facilitate user operations, while for addresses that cannot be completely matched, the user can manually correct them by themselves, ensuring the flexibility of address changes;

[0093] Modification confirmation unit: Based on the address feedback report, it performs a secondary confirmation of the modified content, displays the modified address and the original address, generates the final order address, which serves as the basis for transportation, delivery, and inspection, and generates a final order address table. By showing the comparison before and after the modification, it confirms that the user has accurately corrected the address information, prevents incorrect submission, and thus avoids affecting the subsequent identification of malicious hoarding orders due to ambiguous order addresses.

[0094] Please refer toFigure 3 , the malicious order recognition module includes a personnel information recognition unit, an address frequency statistics unit, and an order details retrieval unit, where:

[0095] Personnel information recognition unit: Based on the final order address table, extract the personnel information related to the address in the order text content, remove duplicate and incomplete information through cleaning, and count the frequently used personnel information through a basic counting algorithm to generate a personal information collation report. Extract personnel information such as name and contact information from different fields of the order, and clean the duplicate personnel information in the same order to ensure the simplicity and integrity of the data. At the same time, count the number of times each personnel information appears within a specified time period to screen out the frequently used personnel information;

[0096] Address frequency statistics unit: Based on the final order address table, use a prefix tree and counting to statistically analyze the usage frequency of each order address, sort the addresses according to the time period and frequency, identify high-frequency addresses and low-frequency addresses, and mark the high-frequency addresses to generate an address frequency statistics table. Similarly, by statistically analyzing the calibrated and accurate order addresses within a specified time period, it is possible to avoid missing some order addresses that are frequently used in a short time due to the influence of fuzzy order addresses;

[0097] Order details retrieval unit: Based on the personal information collation report and the address frequency statistics table, analyze the correlation degree between high-frequency addresses and personnel information through a clustering algorithm, retrieve the orders corresponding to high-frequency addresses and the same personnel information, summarize the specific product information and order time of each order, and generate an order statistics report. After screening out the orders belonging to merchant accounts, analyze the order data through a clustering algorithm, identify and screen out the personnel information and corresponding accounts that frequently place orders in a certain area or multiple areas within a short time, and retrieve the actual item information of the corresponding orders to count information such as the order time, whether it is a popular product, and the order quantity, thereby providing an analysis basis for subsequent judgment of whether there is malicious hoarding behavior.

[0098] Please refer to Figure 2 , the risk prediction module includes an information matching unit, a historical information screening unit, and a risk assessment unit, where:

[0099] Information matching unit: Based on the order statistical report, analyze the patterns between products and order addresses and personnel information within a time period through cosine similarity, analyze the possibility of association, identify potential malicious behaviors, and generate a malicious order analysis report. On the basis of address and recipient information, add the actual ordered product information, vectorize the relevant data of each order, calculate the similarity between different orders through cosine similarity, and identify the behavioral change patterns of users in a short period, such as changes in order frequency and preferences for product selection, etc., to identify potential patterns in user behavior, and then determine whether there is a significant correlation between address, recipient, and product information;

[0100] Historical information screening unit: Based on the malicious order analysis report, screen the historical order data of the corresponding account, analyze the time series of historical orders, identify the normal fluctuation patterns of purchase behaviors, and generate a historical order analysis report. By retrieving the historical order information of the screened account, such as the actual order address, types of purchased products, etc., the historical behavioral patterns can be analyzed, and then their consumption preferences, purchase habits, shopping cycles, etc. can be revealed;

[0101] Risk assessment unit: Based on the historical order analysis report, perform anomaly detection on the order information within the current time period, identify orders that deviate from normal behaviors through classification algorithms, and generate a risk identification report. Compare the current order placement patterns with the historical behavioral patterns, analyze the differences from normal order placement behaviors, so as to determine whether there is malicious hoarding behavior. In summary, by combining semantic recognition to obtain a precisely calibrated order address, malicious hoarders can be effectively prevented from bypassing monitoring through fuzzy addresses, thereby achieving the purpose of maintaining market stability.

[0102] Please refer to Figure 2 , the data processing module includes a log and monitoring unit, a data conversion unit, and an upload and backup unit, where:

[0103] Log and monitoring unit: Based on the risk identification report, collect and monitor the operation logs, error logs, and security logs of the system, monitor the system status in real time, discover potential security issues and performance bottlenecks, and generate a system log report. Through real-time monitoring of the logs, potential security issues and system bottlenecks can be discovered in a timely manner, and then processed in a timely manner to prevent accidents;

[0104] Data conversion unit: Based on the system log report, convert the formats of known data from different sources into formats that meet specific requirements to ensure data consistency, and generate a standardized data set. By converting all system-related data into a unified format, it is convenient for subsequent backup and retrieval;

[0105] Upload Backup Unit: Based on a standardized dataset, it regularly backs up data to the cloud. Through encryption and version control, it ensures the validity and recoverability of the data, generates a system data processing report. By uploading all data to the cloud, data loss can be prevented, and thus when problems occur in the system, it is convenient to provide all the required data information to the administrator to better optimize the overall system.

[0106] The above are only the preferred embodiments of the present invention, and there are no other forms of limitations on the present invention. Any person skilled in the relevant art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An order address recognition system based on semantic recognition technology, characterized in that: include: Text preprocessing module: Based on each text content in the order address, the text content is segmented through rules and statistical algorithms, meaningless words, single characters and special symbols are filtered out, valid text content is retained and corrected, and an order text preprocessing set is generated; Language identification module: Based on the order text preprocessing set, the corresponding language type is determined through the language detection model, and the corresponding analysis is performed at the same time, and the language is translated into the specified type. The texts in different languages ​​are associated through mapping, and the order address is organized into a unified language to generate a basic output text; Semantic recognition module: Based on the basic output text, the deep learning model is used to analyze the vocabulary relationship and long-term and short-term dependencies in the context, understand the semantics of the text content, judge the integrity and validity of the address information, preliminarily identify the missing basic units of administrative divisions in the address structure, and generate an order content recognition report; Address sorting module: Based on the order content recognition report, correct the typos in the text content according to the built-in dictionary, supplement the missing basic units of administrative divisions in the order address, adjust the address structure according to the set order, and generate a standard order address set; Address verification module: Based on the standard order address set, the corresponding address space coordinates are matched through the GIS map tool, the input address structure is compared with the address structure provided by GIS, reverse geocoding is performed, missing parts are inferred, conflicting parts are modified, a complete order address is constructed, and an optimized address information set is generated; User communication module: Based on the address information optimization set, the abnormal composition range of the order address is prompted through the human-computer interaction interface, correction suggestions are provided, and the user is allowed to modify the address and select the recommended address, and a second confirmation is performed after the modification to generate the final order address table; Malicious order identification module: Based on the final order address table, the personnel information of the order address is counted, and the usage frequency corresponding to each order address is calculated, and the high-frequency and low-frequency addresses are identified. According to the high-frequency address and personnel information, the corresponding order details are retrieved, and the goods purchased in each batch of orders are checked to generate an order statistics report; Risk prediction module: Based on the order statistics report, the correlation between the order address, personnel information and purchased goods is analyzed through machine learning and data analysis technology, and abnormal orders are screened out by combining anomaly detection, time series analysis, classification algorithm and clustering algorithm, the probability of malicious hoarding is evaluated, and a risk identification report is generated; Data processing module: Based on the risk identification report, collect system operation logs, convert the format of known information and record backup, and generate a system data processing report.

2. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The text preprocessing module includes a text segmentation unit, a word deletion unit and a specification correction unit, wherein: Text segmentation unit: Based on the input text content, the text content is segmented in multiple steps and dimensions through rules and statistical algorithms, irregular order addresses are divided into multiple basic texts, and a text segmentation report is generated; Word deletion unit: based on the text segmentation report, the text content is cleaned by regular expressions, stem extraction and word form restoration to reduce redundancy and generate a text cleaning report; Specification correction unit: Based on the text cleaning report, letter conversion and correction are performed through a natural language processing model to maintain the uniformity of text content and generate an order text preprocessing set.

3. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The language type identification module includes a language detection unit, a parsing and fusion unit, and a language conversion and output unit, wherein: A language detection unit: based on the order text preprocessing set, extracting text features through Langid, including character frequency and letter sequence combination, counting the probability of occurrence of specific words or word combinations in each language, and using the naive Bayes algorithm to generate a classification model for each language, screening out the corresponding text language, determining the required translation model, and generating a text language screening report; Parsing and fusion unit: based on the text language screening report, the text is translated through the corresponding translation model, and the machine learning model is used with the mapping table to match different text languages ​​to generate a text content association set; Language conversion output unit: based on the text content association set, compares and converts the different input text languages ​​with the standard format, performs cross-language text integration, and generates basic output text.

4. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The semantic recognition module includes a database construction unit, a model training unit and a semantic understanding unit, wherein: Database construction unit: Based on the basic output text, the data is mapped to a high-dimensional space through a deep learning model, the stored standard order text is vectorized into sentences and words, and the text is embedded in the annotated space to wait for subsequent matching, thereby generating a basic data set; Model training unit: Based on the basic data set, in combination with experimental orders with correct and incorrect address information and historical actual orders, the learning model is trained through a supervised learning algorithm to improve the ability to predict missing content in the text, and the model parameters are adjusted through cross-validation, accuracy evaluation and loss function optimization to generate a model training library; Semantic understanding unit: Based on the model training library, it matches highly similar entries, analyzes the input text content through natural language processing technology, extracts relevant information related to the address, determines the completeness and consistency of the information, supplements it with context, and generates an order content recognition report.

5. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The address arrangement module includes a spelling correction unit, an element supplement unit and a standardization arrangement unit, wherein: Spelling correction unit: Based on the order content recognition report, the words in the order content are matched by a built-in dictionary, the inaccurately spelled words are identified, and the incorrect spellings are replaced with correct spellings according to preset dictionary rules to generate an address correction report; Element supplement unit: Based on the address correction report, the missing administrative division units in the text content are identified by using named entity recognition technology, and the missing administrative division units are inferred and filled according to the specific name of the address and context information to generate an address supplement report; Standardization sorting unit: Based on the address supplementary report, according to the hierarchical relationship constraints, the address composition order is adjusted according to the set administrative division tree, and the separators in the address are unified into a standard form to generate a standard order address set.

6. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The address checking module includes a geographic coding unit, a matching and comparison unit, and a depth optimization unit, wherein: Geocoding unit: Based on the standard order address set, the input text content is parsed to extract the relevant address information of the street, city, and country, and the given standard order address is converted into the corresponding geographic space coordinates through the geocoding algorithm in combination with the map data in the GIS database to generate a coordinate conversion report; Matching and comparison unit: Based on the coordinate conversion report, the order address is matched and compared with the address provided by the GIS standard address library through a string matching algorithm and geographic location similarity analysis, the differences and conflicts between the two are checked, the missing information, redundant parts and spelling errors still existing in the order address are identified, and marked, and an address matching report is generated; Deep optimization unit: Based on the address matching report, reverse encoding is performed according to the coordinate points through a machine learning algorithm, and the detailed text content of the order address is reversed and filled in, and the unmatched place names are corrected to generate an optimized set of address information.

7. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The user communication module includes an error prompt unit, a user feedback unit and a modification confirmation unit, wherein: Error prompt unit: Based on the address information optimization set, the abnormal areas and fields of the order address are highlighted through the UI interface, a clear correction range description is provided, the possibility and cause of the error is informed, the user is assisted in identifying the problem and making corresponding corrections, and a problem address prompt report is generated; User feedback unit: Based on the problem address prompt report, the user is provided with correction suggestions through the UI interface, and the user manually modifies the address for the correction suggestions that are not completely accurate, and generates an address feedback report; Modification confirmation unit: Based on the address feedback report, the modification content is reconfirmed, the modified address and the original address are displayed, the final order address is generated as a basis for transportation, delivery and inspection, and a final order address table is generated.

8. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The malicious order identification module includes a personnel information identification unit, an address frequency statistics unit and an order details retrieval unit, wherein: Personnel information identification unit: based on the final order address table, extracts personnel information related to the address in the order text content, removes duplicate and incomplete information through cleaning, and uses a basic counting algorithm to count frequently used personnel information to generate a personal information sorting report; Address frequency statistics unit: based on the final order address table, the usage frequency of each order address is counted by prefix tree and counting, the addresses are sorted according to time period and frequency, high-frequency addresses and low-frequency addresses are identified, and high-frequency addresses are marked to generate an address frequency statistics table; Order details retrieval unit: Based on the personal information compilation report and address frequency statistics table, the correlation between high-frequency addresses and personnel information is analyzed through a clustering algorithm, and the orders corresponding to high-frequency addresses and the same personnel information are retrieved, and the specific product information and order time of each order are summarized to generate an order statistics report.

9. The order address recognition system based on semantic recognition technology according to claim 1 is characterized in that: The risk prediction module includes an information matching unit, a historical information screening unit and a risk assessment unit, wherein: Information matching unit: Based on the order statistics report, the information matching unit analyzes the rules between the goods, order addresses and personnel information within a time period through cosine similarity, analyzes the possibility of association, identifies potential malicious behaviors, and generates a malicious order analysis report; A historical information screening unit: based on the malicious order analysis report, screening the historical order data of the corresponding account, analyzing the time series of the historical orders, identifying the normal fluctuation pattern of the purchase behavior, and generating a historical order analysis report; Risk assessment unit: Based on the historical order analysis report, perform anomaly detection on the order information in the current time period, identify orders that deviate from normal behavior through a classification algorithm, and generate a risk identification report.

10. The order address recognition system based on semantic recognition technology according to claim 1, characterized in that: The data processing module includes a log and monitoring unit, a data conversion unit and an upload backup unit, wherein: Log and monitoring unit: based on the risk identification report, collect and monitor the system's operation log, error log and security log, monitor the system status in real time, discover potential security issues and performance bottlenecks, and generate a system log report; Data conversion unit: based on the system log report, convert the formats of known data from different sources into formats that meet specific requirements, ensure data consistency, and generate standardized data sets; Upload backup unit: Based on the standardized data set, it is backed up to the cloud regularly, and the validity and recoverability of the data are ensured through encryption and version control, and a system data processing report is generated.

Images