A Smart Identification Method for Street-Front Shops Based on Improved TF-IDF

Abstract

This invention discloses an intelligent identification method for street-front shops based on an improved TF-IDF model, belonging to the field of intelligent shop identification technology. The method includes: extracting and normalizing internal customer information from the operator to obtain a local customer pool; performing a three-dimensional matching between a clean shop pool and the local customer pool to obtain a first pool of matched shops and a second pool of unmatched shops; forming text pairs using corresponding fields from the second shop pool and the local customer pool, calculating semantic similarity using an improved TF-IDF model, and identifying shops with similarity greater than a threshold as model-matched local shops; merging the first shop pool with the model-matched local shops to form the final local shop pool, and removing them from the clean shop pool to obtain a pool of shops from other networks; adding industry and product tags to the final local and other network shop pools to generate a shop profile library and outputting it to the marketing end. This invention can improve the accuracy of shop identification, reduce the cost of manual visits, and achieve one-time separation of local and other network shops for precise marketing.

Description

Technical Field

[0001] This invention relates to the field of intelligent shop recognition technology, and in particular to an intelligent recognition method for street-front shops based on improved TF-IDF. Background Technology

[0002] With the increasing demand from telecom operators for refined operations of street-front shops, it is necessary to have a complete inventory of all shops and accurately identify which are existing customers of the network and which are potential customers of other networks. Current technologies primarily employ the following solutions: They generally rely on a one-time distribution by the group company or crawling from a single map, with an update frequency as low as once a year, making it difficult to cover changes in shops that open or close rapidly.

[0003] The distributed or crawled results contain a large number of non-commercial objects such as road ancillary facilities, government agencies, and commercial and residential properties; the lack of unified cleaning rules leads to high noise in the subsequent matching process.

[0004] Simply using precise equivalence matching and simple fuzzy matching of customer names, addresses, or contact numbers to associate shops with customers cannot solve textual discrepancies such as identical names at different addresses, misspellings, and missing address levels. On-site manual visits are still required for confirmation, which is costly and inefficient.

[0005] Traditional TF-IDF focuses on the importance of words that appear less frequently in all documents. It does not give enough weight to distinguishing terms that appear frequently in the customer group but are rare in other groups, resulting in distortion of the store-customer name / address similarity calculation and further reducing the accuracy of automatic recognition.

[0006] Therefore, existing technologies suffer from drawbacks such as narrow data coverage, slow updates, weak cleaning, and coarse matching algorithms. There is an urgent need for an intelligent recognition method that can acquire data from multiple sources at high frequency, effectively filter non-shop objects, and achieve high-precision semantic matching by means of an improved text weight model. Summary of the Invention

[0007] The technical problem to be solved by this invention is to address the shortcomings of existing technologies, specifically by providing an intelligent identification method for street-front shops based on an improved TF-IDF, as detailed below: 1) In a first aspect, the present invention provides a method for intelligent identification of street-front shops based on improved TF-IDF, the specific technical solution of which is as follows: S1, extract customer information from within the operator and normalize it using standard addresses to obtain the customer pool of this network; S2, perform a three-dimensional matching process between the pre-constructed clean shop pool and the customer pool of this network, which consists of telephone, name and address, to obtain a first shop pool consisting of matched shops and a second shop pool consisting of unmatched shops; S3, using the shop name or shop address in the second shop pool as a basis, form text pairs with the customer name or line management standard address in the customer pool of this network respectively, use the improved TF-IDF model to calculate the semantic similarity of any text pair, and confirm the shops with semantic similarity greater than the threshold as the shops that match the model of this network. S4, merge the first shop pool and the model-matched local shops to form the final local shops, and deduplicate the final local shops with the clean shop pool to obtain the heterogeneous shop pool. S5, add industry and product tags to the final local online store and the external online store pool respectively, generate a store profile library and output it to the marketing end.

[0008] The beneficial effects of the intelligent identification method for street-front shops based on improved TF-IDF provided by this invention are as follows: S1 standardizes the local customer pool, ensuring customer addresses have unified coordinates, allowing direct comparison with online store addresses and reducing missed matches due to address format differences. S2's three-dimensional precise matching identifies high-confidence local stores when any one of the phone number, name, or address is identical, quickly locking in existing customers and reducing subsequent computational load. S3 uses an improved TF-IDF model for the second store pool, leveraging frequency to increase the weight of store feature words, ensuring high similarity even with abbreviations, aliases, or reversed order of names or addresses. S4 merges the first store pool with the model matching results and removes duplicates in one go, ensuring no duplicate local stores and immediately generating a cross-network store pool, achieving one-time separation of local and cross-network stores. S5, with added industry and product tags, creates a store profile library that can be directly pushed to the marketing end, allowing frontline staff to maintain existing customers or acquire new customers from cross-network stores based on the profiles, eliminating the need for door-to-door verification, thereby improving operational efficiency and promoting business development.

[0009] Based on the above solution, the present invention can be further improved as follows.

[0010] Furthermore, the construction process of the pre-built clean shop pool includes: Raw shop data is obtained from multiple sources of geographic and lifestyle service platforms and aggregated to generate a raw shop pool; The original shop pool is subjected to standardization, deduplication, and verification processes in sequence to obtain the processing results. Invalid information is filtered out from the processing results to obtain a pre-constructed clean shop pool.

[0011] Furthermore, the process of calculating the semantic similarity of any text pair using the improved TF-IDF model is as follows: Generate word frequency vectors for the two texts in the text pair, weight each word t in the vector according to the first formula, calculate the cosine similarity between the two vectors, and the obtained cosine value is the semantic similarity. The first formula is as follows: IDF = log[(Nk) / m]; Where m is the number of texts from our website's clients that contain term t, k is the number of texts from non-our website's clients that contain term t, N is the total number of texts, and IDF is the weighted processing result.

[0012] Furthermore, it also includes: Using the first pool of shops as positive samples and the second pool of shops as negative samples, a classifier is trained using a genetic algorithm. The classifier is then used to verify the model's matching with shops on this website. Models that pass the verification are retained and S4 is executed.

[0013] 2) Secondly, the present invention also provides an intelligent identification system for street-front shops based on improved TF-IDF, the specific technical solution of which is as follows: The acquisition module is used to: extract customer information from within the operator and process it through standard address normalization to obtain the customer pool of this network; The matching module is used to: perform three-dimensional matching of the pre-built clean store pool with the customer pool of this network, namely, telephone, name and address, to obtain a first store pool consisting of matched stores and a second store pool consisting of unmatched stores; The similarity calculation module is used to: take the shop name or shop address in the second shop pool as the basis, form text pairs with the customer name or line management standard address in the customer pool of this network respectively, use the improved TF-IDF model to calculate the semantic similarity of any text pair, and confirm the shops with semantic similarity greater than the threshold as the shops that match the model of this network. The merging module is used to: merge the first shop pool and the model-matched local shops to form the final local shops, and deduplicate the final local shops with the clean shop pool to obtain the heterogeneous shop pool; The identification module is used to: add industry and product tags to the final local online store and the external online store pool respectively, generate a store profile library and output it to the marketing end.

[0014] Based on the above solution, the present invention can be further improved as follows.

[0015] Furthermore, the construction process of the pre-built clean shop pool includes: Raw shop data is obtained from multiple sources of geographic and lifestyle service platforms and aggregated to generate a raw shop pool; The original shop pool is subjected to standardization, deduplication, and verification processes in sequence to obtain the processing results. Invalid information is filtered out from the processing results to obtain a pre-constructed clean shop pool.

[0016] Furthermore, the process of calculating the semantic similarity of any text pair using the improved TF-IDF model is as follows: Generate word frequency vectors for the two texts in the text pair, weight each word t in the vector according to the first formula, calculate the cosine similarity between the two vectors, and the obtained cosine value is the semantic similarity. The first formula is as follows: IDF = log[(Nk) / m]; Where m is the number of texts from our website's clients that contain term t, k is the number of texts from non-our website's clients that contain term t, N is the total number of texts, and IDF is the weighted processing result.

[0017] Furthermore, it also includes: Using the first pool of shops as positive samples and the second pool of shops as negative samples, a classifier is trained using a genetic algorithm. The classifier is then used to verify the model's matching with shops on this network. The models that pass the verification are retained, and the merging module is executed.

[0018] 3) In a third aspect, the present invention also provides an electronic device, the electronic device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the electronic device to perform any of the above methods.

[0019] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0020] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0021] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an intelligent identification method for street-front shops based on an improved TF-IDF according to an embodiment of the present invention. Figure 2This is a schematic diagram of the model architecture of a street-front shop intelligent identification method based on improved TF-IDF according to an embodiment of the present invention; Figure 3 This is a structural framework diagram of an electronic device according to the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0023] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for intelligent identification of street-front shops based on improved TF-IDF, which includes the following steps: S1, extract customer information from within the operator and normalize it using standard addresses to obtain the customer pool of this network; S2, perform a three-dimensional matching process between the pre-constructed clean shop pool and the customer pool of this network, which consists of telephone, name and address, to obtain a first shop pool consisting of matched shops and a second shop pool consisting of unmatched shops; S3, using the shop name or shop address in the second shop pool as a basis, form text pairs with the customer name or line management standard address in the customer pool of this network respectively, use the improved TF-IDF model to calculate the semantic similarity of any text pair, and confirm the shops with semantic similarity greater than the threshold as the shops that match the model of this network. S4, merge the first shop pool and the model-matched local shops to form the final local shops, and deduplicate the final local shops with the clean shop pool to obtain the heterogeneous shop pool. S5, add industry and product tags to the final local online store and the external online store pool respectively, generate a store profile library and output it to the marketing end.

[0024] The beneficial effects of the intelligent identification method for street-front shops based on improved TF-IDF provided by this invention are as follows: S1 standardizes the local customer pool, ensuring customer addresses have unified coordinates, allowing direct comparison with online store addresses and reducing missed matches due to address format differences. S2's three-dimensional precise matching identifies high-confidence local stores when any one of the phone number, name, or address is identical, quickly locking in existing customers and reducing subsequent computational load. S3 uses an improved TF-IDF model for the second store pool, leveraging frequency to increase the weight of store feature words, ensuring high similarity even with abbreviations, aliases, or reversed order of names or addresses. S4 merges the first store pool with the model matching results and removes duplicates in one go, ensuring no duplicate local stores and immediately generating a cross-network store pool, achieving one-time separation of local and cross-network stores. S5, with added industry and product tags, creates a store profile library that can be directly pushed to the marketing end, allowing frontline staff to maintain existing customers or acquire new customers from cross-network stores based on the profiles, eliminating the need for door-to-door verification, thereby improving operational efficiency and promoting business development.

[0025] In another embodiment of this solution, corporate customer information is extracted from the operator's internal B-domain business system and O-domain building system. The fields are unified as customer name, customer contact number, and customer address. The original address is normalized using the number line management standard address database, and missing administrative divisions, roads, and house numbers are filled in and the coordinate system is unified to form the network's customer pool. Here, "internal to the operator" refers to the telecom operator's B-domain and O-domain information systems, and "number line management standard address" is the operator's standardized address database used for broadband number line resource management.

[0026] Among them, the B domain business system is the system in telecommunications operators that manages users and money, also known as the business support system (BSS). This system includes all data related to marketing, billing, and customer service, such as customer information, orders, packages, bills, and payment records. The O-Domain Building System is part of the Operation Support System (OSS). It is responsible for managing network resources and physical addresses, recording the addresses and coverage methods of network resources such as broadband lines, equipment rooms, pipelines, buildings, and optical splitters. In addition, the standard address for line management is a standardized address database, which is included in the O-Domain Building System and is used to convert user-submitted addresses into resource addresses that can be constructed and dispatched.

[0027] In another embodiment of this solution, a pre-built clean shop pool is obtained. The pre-built clean shop pool has been collected in the early stage through a multi-source map platform and business system and has undergone standardization, deduplication and invalid list filtering. The contact phone number, shop name and shop address of each shop in the pre-built clean shop pool are matched precisely with the corresponding fields of the customer pool of this network in terms of phone number dimension, name dimension and address dimension. If any dimension is completely consistent, it is considered a matched shop and is included in the first shop pool. The remaining unmatched shops constitute the second shop pool.

[0028] It should be further explained that the process of precise equal-value matching is as follows: the contact phone number, shop name, and shop address of each shop in the clean shop pool are compared with the contact phone number, customer name, and customer contact address of the customers in the network's customer pool to determine if they are completely equal in the same field. If any string in the phone number, name, or address dimension is completely identical, the shop is immediately added to the first shop pool, and the remaining shops that do not have completely identical items are assigned to the second shop pool.

[0029] In another embodiment of this solution, using the second shop pool as a benchmark, the shop name and customer name of each shop in the second shop pool are combined into a name text pair, and the shop address and the line management standard address are combined into an address text pair. The name text pair and the address text pair together constitute a text pair. For each text pair, an improved TF-IDF model is used to calculate semantic similarity. The improved TF-IDF model reweights IDF using intra-class frequency m and out-of-class frequency k to highlight shop feature words. When the obtained cosine similarity is greater than a preset threshold σ, the shop is confirmed as a shop that matches the model on this network. The threshold is a similarity threshold set empirically.

[0030] In another embodiment of this solution, the first shop pool is merged with the shops in the model matching network. During the merging process, duplicates are removed according to the unique identifier of the shop to form the final shop in the network without duplicates. The final shop in the network is removed from the clean shop pool, and the remaining part directly constitutes the shop pool of the other network, realizing the one-time separation of the network and the other network.

[0031] In another embodiment of this solution, the final pool of local and external online stores are connected to Meituan category fields and Qichacha industry fields, respectively, and industry tags and product tags are automatically added to generate a store profile library with tags. The profile library is then pushed to the existing customer retention system and potential new customer acquisition system on the marketing end in real time through the interface to support subsequent precise operations.

[0032] Furthermore, the construction process of the pre-built clean shop pool includes: Raw shop data is obtained from multiple sources of geographic and lifestyle service platforms and aggregated to generate a raw shop pool; The original shop pool is subjected to standardization, deduplication, and verification processes in sequence to obtain the processing results. Invalid information is filtered out from the processing results to obtain a pre-constructed clean shop pool.

[0033] Furthermore, the process of calculating the semantic similarity of any text pair using the improved TF-IDF model is as follows: Generate word frequency vectors for the two texts in the text pair, weight each word t in the vector according to the first formula, calculate the cosine similarity between the two vectors, and the obtained cosine value is the semantic similarity. The first formula is as follows: IDF = log[(Nk) / m]; Where m is the number of texts from our website's clients that contain term t, k is the number of texts from non-our website's clients that contain term t, N is the total number of texts, and IDF is the weighted processing result.

[0034] Furthermore, it also includes: Using the first pool of shops as positive samples and the second pool of shops as negative samples, a classifier is trained using a genetic algorithm. The classifier is then used to verify the model's matching with shops on this website. Models that pass the verification are retained and S4 is executed.

[0035] Example 1, Step 1, Expanding the sources and frequency of store data: The data acquisition frequency is increased from once a year to three times a year; Step 2, Strengthen the verification and filtering of store data quality: 1) Integrate multi-source store data, including data standardization, deduplication, and data verification; 2) Filter invalid information (non-commercial), including road ancillary facilities, place name and address information, scenic spots, public facilities, commercial and residential properties, transportation facilities and services, events and activities, indoor facilities, access facilities, government agencies and social organizations, as well as schools, telecommunications business halls, shared equipment, ATMs, banks, hospitals, gas stations, CNG stations, funeral facilities, nucleic acid testing, archives, etc. in the secondary categories.

[0036] Step 3: Enhance the intelligent identification capabilities of street-front shops via the internet: The data of online shops crawled from a map is associated with the standard addresses managed by the number management system through contact phone number, shop name, and shop address. It is also associated with actual customers in the system. By using an improved TF-IDF model to calculate the similarity between online shops and system shops in multiple dimensions (name, address, phone number, latitude and longitude), the ability to identify online street shops is enhanced, and online street shops are more accurately identified.

[0037] Introduction to the standard TF-IDF model The main idea of ​​TF-IDF is that the weight of a feature word in a document is inversely proportional to the frequency of the feature word in the document and the number of documents containing that feature word. TF represents the frequency of the feature word in document D, and IDF represents the number of documents containing the feature word across all documents.

[0038] Commonly used calculation methods are as follows: Where m represents the number of times the feature word appears in document i, and M represents the total number of words in document i.

[0039] Where N is the total number of documents, and n is the total number of documents containing a certain feature word.

[0040] The main idea of ​​IDF is that the fewer documents containing term t (i.e., the smaller n is), the larger the IDF, indicating that term t has good class distinguishing ability. If the number of documents containing term t in a certain class C is m, and the total number of documents containing t in other classes is k, then obviously the total number of documents containing t is n = m + k. When m is large, n is also large, and the IDF value obtained according to the IDF formula will be small, indicating that term t has weak class distinguishing ability. However, in reality, if a term appears frequently in documents of a certain class, it means that the term can well represent the characteristics of the text of that class. Such terms should be given higher weights and selected as feature words for that class of text to distinguish it from documents of other classes. This is the shortcoming of IDF.

[0041] The improved TFIDF model in this solution: An improvement to IDF is proposed. Let the total number of documents be N, the number of documents containing term t be n, and the number of documents containing term t in a certain class Ci be m. Then, t in class Ci is: If the number of documents containing term t is large in a certain class Ci, while the number of documents containing term t is small in other classes, then t can represent the characteristics of the text in class Ci and has good class distinguishing ability. If the number of documents containing term t other than class Ci is k, then the formula is transformed into the following form: The relationship between the value of IDF and m and k is as follows: ; but Since m1>0, m2>0, and k>0, then f(m1)-f(m2)>0. It increases as m increases. Clearly, It decreases as k increases. In other words, the IDF value increases as m increases and decreases as k increases. If a class Ci has a large number of documents containing term t, while other classes have a small number of documents containing term t, then t can represent the characteristics of class Ci.

[0042] exist Figure 2 In this context, each class has a classifier called a class classifier. To determine the category of a document, each class classifier determines whether the document belongs to the class it represents.

[0043] In the training text for each class, a genetic algorithm is used to train a class classifier (class text vector). Genetic algorithms have excellent optimization capabilities; leveraging this characteristic, a text vector representing the class is trained from the training documents. Once the class classifier is trained, to determine if a test document belongs to that class, simply compare the similarity between the test document and the class classifier.

[0044] The improved TF-IDF is used to measure the weight of the terms in the training documents of each class, and the terms are sorted according to their weights. The K terms with the largest weights are selected from each document.

[0045] The selected terms from each training text are placed into a set A after removing duplicates. All terms in A can form a string like a1, a2, a3...an, where n is the number of elements in set A and ai represents the i-th element in A.

[0046] If `ai` appears in the document, its corresponding position is replaced with 1; otherwise, it is replaced with 0. The document can be represented as a string: 101...1...0. To determine the similarity between two texts, we generally determine the similarity between their text vectors. Vector similarity can be measured by the cosine of the angle between the two vectors. The larger the value, the higher the similarity. Let the two text vectors... and Then, the cosine of the angle between the two vectors is: ; If cos(d) i ,d j If )>σ, then d is considered to be i and d j If the similarity is found, then the similarity is found; otherwise, the similarity is not found. Here, σ is a similarity threshold, which is also an empirical value.

[0047] The training texts for each class form an initial population. A class text vector is trained in the training texts for each class as a class classifier. Once the class text vector is trained, the test document is judged to belong to this class based on the cosine of the angle between the test document and the class text vector. If the cosine is greater than σ, the text belongs to this class; otherwise, it does not belong to this class.

[0048] Given text d and terms in set A, when ai appears in text d, the corresponding bits of the strings a1, a2, a3, ..., ai..., an are set to 1; otherwise, they are set to 0. Thus, based on whether a term in A appears in d, d can be represented as a binary string of 1010001...0101. The string length is n, which is the number of elements in A.

[0049] When training the class classifier, training texts from all classes are used to train the fitness of the class text vector, which determines whether it can correctly identify text of its own class and accurately distinguish text of other classes. Assuming C represents a text vector d in class C, the fitness of d can be expressed as: Where T+ represents the number of texts that belong to Ci and can be correctly classified into Ci using d, T- represents the number of texts that belong to Ci but are not correctly classified into Ci, F+ represents the number of texts that do not belong to Ci but are classified into Ci, and F- represents the number of texts that do not belong to Ci and are not classified into Ci.

[0050] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, and these situations are also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0051] This invention also provides an intelligent identification system for street-front shops based on improved TF-IDF, the specific technical solution of which is as follows: The acquisition module is used to: extract customer information from within the operator and process it through standard address normalization to obtain the customer pool of this network; The matching module is used to: perform three-dimensional matching of the pre-built clean store pool with the customer pool of this network, namely, telephone, name and address, to obtain a first store pool consisting of matched stores and a second store pool consisting of unmatched stores; The similarity calculation module is used to: take the shop name or shop address in the second shop pool as the basis, form text pairs with the customer name or line management standard address in the customer pool of this network respectively, use the improved TF-IDF model to calculate the semantic similarity of any text pair, and confirm the shops with semantic similarity greater than the threshold as the shops that match the model of this network. The merging module is used to: merge the first shop pool and the model-matched local shops to form the final local shops, and deduplicate the final local shops with the clean shop pool to obtain the heterogeneous shop pool; The identification module is used to: add industry and product tags to the final local online store and the external online store pool respectively, generate a store profile library and output it to the marketing end.

[0052] It should be noted that the beneficial effects of the improved TF-IDF-based intelligent identification system for street-front shops provided in the above embodiments are the same as those of the improved TF-IDF-based intelligent identification method for street-front shops, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0053] like Figure 3 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320 coupled to a memory 310. The memory 310 stores at least one computer program 330, which is loaded and executed by the processor 320 to enable the electronic device 300 to implement any of the above-mentioned methods. Specifically: The electronic device 300 can vary considerably due to differences in configuration or performance. It may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. The memories 310 store at least one computer program 330, which is loaded and executed by the processors 320 to enable the electronic device 300 to implement the improved TF-IDF-based intelligent identification method for street-front shops provided in the above embodiments. Of course, the electronic device 300 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. It may also include other components for implementing device functions, which will not be elaborated upon here.

[0054] An embodiment of the present invention provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods.

[0055] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0056] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the methods described above.

[0057] It should be noted that the terms "first" and "second" in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0058] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0059] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0060] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An improved TF-IDF based method for intelligent recognition of street shops, characterized in that, The method comprises the following steps: S1, extracting customer information within an operator and performing standard address normalization processing to obtain a customer pool in the network; S2, performing telephone-name-address three-dimensional matching processing on the pre-constructed clean shop pool and the customer pool in the network to obtain a first shop pool composed of matched shops and a second shop pool composed of unmatched shops; S3, taking the shop name or shop address in the second shop pool as a reference, respectively forming a text pair with the customer name or number line management standard address in the customer pool in the network, calculating the semantic similarity of any text pair by using an improved TF-IDF model, and confirming the shop with a semantic similarity greater than a threshold value as a model-matched shop in the network; S4, performing merging processing on the first shop pool and the model-matched shop in the network to form a final shop in the network, performing deduplication processing on the final shop in the network and the clean shop pool to obtain a shop pool in a different network; S5, respectively adding industry and product labels to the final shop in the network and the shop pool in the different network, generating a shop portrait library and outputting to a marketing end. 2.The improved TF-IDF based method for street shop intelligent recognition according to claim 1, characterized in that, The construction process of the pre-constructed clean shop pool comprises: obtaining original shop data from multiple source geographic and life service platforms, and generating an original shop pool by collection; performing standardization processing, deduplication processing and verification processing on the original shop pool in sequence to obtain a processing result, and performing invalid information filtering processing on the processing result to obtain a pre-constructed clean shop pool. 3.The improved TF-IDF based method for street shop intelligent recognition according to claim 1, characterized in that, The process of calculating the semantic similarity of any text pair by using the improved TF-IDF model comprises: generating a word frequency vector for each of the two texts in the text pair, performing weighting processing on each word t in the vector according to a first formula, calculating the cosine similarity of the two vectors, and the obtained cosine value is the semantic similarity; the first formula is specifically: IDF=log[(N-k) / m]; wherein m is the number of non-network customer texts containing the word t, k is the number of non-network customer texts containing the word t, N is the total number of texts, and IDF is the weighting processing result. 4.The improved TF-IDF based method for street shop intelligent recognition according to claim 1, characterized in that, Further comprising: taking the first shop pool as a positive sample and the second shop pool as a negative sample, training a class classifier by using a genetic algorithm, verifying the model-matched shop in the network by using the class classifier, retaining the model-matched shop in the network that passes the verification, and performing S4.

5. An improved TF-IDF based street shop intelligent identification system, characterized in that, The method comprises the following steps: The acquisition module is configured to extract customer information within an operator and perform standard address normalization processing to obtain a customer pool in the network; The matching module is configured to perform telephone-name-address three-dimensional matching processing on the pre-constructed clean shop pool and the customer pool in the network to obtain a first shop pool composed of matched shops and a second shop pool composed of unmatched shops; The similarity calculation module is configured to take the shop name or shop address in the second shop pool as a reference, respectively form a text pair with the customer name or number line management standard address in the customer pool in the network, calculate the semantic similarity of any text pair by using an improved TF-IDF model, and confirm the shop with a semantic similarity greater than a threshold value as a model-matched shop in the network; The merging module is used to: merge the first shop pool and the model-matched local shops to form the final local shops, and deduplicate the final local shops with the clean shop pool to obtain the heterogeneous shop pool; The identification module is used to: add industry and product tags to the final local online store and the external online store pool respectively, generate a store profile library and output it to the marketing end. 6.The improved TF-IDF based street shop intelligent recognition system according to claim 5, wherein, The process of constructing the pre-built clean shop pool includes: Raw shop data is obtained from multiple sources of geographic and lifestyle service platforms and aggregated to generate a raw shop pool; The original shop pool is subjected to standardization, deduplication, and verification processes in sequence to obtain the processing results. Invalid information is filtered out from the processing results to obtain a pre-constructed clean shop pool.

7. The improved TF-IDF based street shop intelligent recognition system according to claim 5, wherein, The process of calculating the semantic similarity of any text pair using the improved TF-IDF model is as follows: Generate word frequency vectors for the two texts in the text pair, weight each word t in the vector according to the first formula, calculate the cosine similarity between the two vectors, and the obtained cosine value is the semantic similarity. The first formula is as follows: IDF = log[(Nk) / m]; Where m is the number of texts from our website's clients that contain term t, k is the number of texts from non-our website's clients that contain term t, N is the total number of texts, and IDF is the weighted processing result. 8.The improved TF-IDF based street shop intelligent recognition system according to claim 5, wherein, Also includes: The verification module is used to: use the first pool of shops as positive samples and the second pool of shops as negative samples, train a classifier using a genetic algorithm, verify the model matching the shops on this network using the classifier, retain the model matching the shops on this network that pass the verification, and execute the merging module.

9. An electronic device, comprising: The electronic device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to perform the method as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to perform the method as described in any one of claims 1 to 4.

Images