Residential quarter affiliation identification method and device based on multi-source data and machine learning

Abstract

The invention provides a residential quarter affiliation identification method and device based on multi-source data and machine learning. The method comprises the following steps: constructing a residential quarter feature library through broadband account opening samples; obtaining signaling data of a user to be identified, and determining a candidate cell set; obtaining a user behavior feature according to the signaling data of the user to be identified, and obtaining a cell feature fingerprint of a first candidate cell according to a cell feature library, and inputting the user behavior characteristics, the cell characteristic fingerprints and the base station matching degree characteristics of the to-be-identified user and the first candidate cell into a pre-trained residential cell classification model as input characteristic vectors to obtain an identification result of the first candidate cell, and sequentially obtaining identification results of all candidate cells in a candidate cell set, obtaining a preliminary cell affiliation result; and obtaining a final residential quarter of the to-be-identified user based on the preliminary residential quarter affiliation result of the family members of the to-be-identified user. According to the invention, the self-adaptive capability and the identification precision of the identification method can be obviously improved.

Description

Technical Field

[0001] This invention relates to the field of communication information processing technology, and in particular to a method and apparatus for identifying residential community affiliation based on multi-source data and machine learning. Background Technology

[0002] In telecommunications marketing, accurately identifying users' residential communities is crucial for expanding the home market, targeted marketing, and urban planning. Existing methods largely rely on cell tower positioning, using third-party mapping tools to obtain community boundaries and then mapping the cell towers to the communities. This approach has the following problems: 1. The base station has a wide coverage area, which can easily lead to users in neighboring cells being mistaken for users in this cell.

[0003] 2. Relying on web crawlers to obtain cell boundaries results in untimely data updates and low boundary accuracy.

[0004] 3. Matching based solely on permanent base stations, without integrating multi-dimensional data such as user behavior characteristics and family relationships, results in significant identification errors.

[0005] Based on this, a residential cell identification scheme based on broadband data and base station matching, published in patent CN117440317A, determines the cell to which a broadband user belongs by using the user's installation address. It then aggregates the nighttime signaling data of these users and filters out one or more coverage base stations corresponding to that cell. For the user to be identified, its nighttime signaling data is analyzed, and its nighttime base stations are filtered out based on the duration and number of days of stay. The user's nighttime base station is matched with the coverage base station database of each cell using preset rules to obtain candidate cells. Finally, preset conditions are used to filter out the final residential cell from the candidate cells. Although the above-mentioned prior art eliminates the dependence on third-party maps, it is essentially still a method based on hard-coded rules and shallow statistics, which has the following inherent drawbacks: the above-mentioned cell identification scheme suffers from poor adaptability and limited identification accuracy due to the setting of rules and the use of features. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides a method and apparatus for identifying residential community affiliation based on multi-source data and machine learning, which improves the adaptability and accuracy of the identification method.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for identifying residential community affiliation based on multi-source data and machine learning, comprising: Step S1: Construct a community feature database using broadband account opening samples; Step S2: Obtain the signaling data of the user to be identified, and determine the candidate cell set based on the signaling data of the user to be identified; Step S3: Based on the user behavior features obtained from the signaling data of the user to be identified, and the cell feature fingerprint of the first candidate cell obtained from the cell feature library, the user behavior features, the cell feature fingerprint, and the base station matching degree features between the user to be identified and the first candidate cell are input as input feature vectors into the pre-trained residential cell classification model to obtain the identification result of the first candidate cell, and then obtain the identification results of all candidate cells in the candidate cell set in sequence to obtain the preliminary cell attribution result; Step S4: Obtain the final residential community of the user to be identified based on the preliminary community affiliation results of the family members of the user to be identified.

[0008] The beneficial effects of the present invention are as follows: The present invention uses the base station matching degree features of the user to be identified and the first candidate cell, the user behavior features of the user to be identified, and the cell feature fingerprint of the first candidate cell as multi-source data and inputs them into the pre-trained residential cell classification model as input feature vector, thereby significantly improving the adaptive ability and recognition accuracy of the identification method.

[0009] Optionally, step S1 includes: Step S11: Construct a community sample library using broadband account opening samples; Step S12: Construct an initial feature set based on the signaling data of all broadband users in each cell in the cell sample library. Input the initial feature sets of all cells into the random forest model to evaluate the feature importance of each cell. Take the top N features with the highest importance scores in each cell as cell feature fingerprints. Construct a cell feature library from the cell feature fingerprints of all cells, where N ≥ 2 and is a positive integer.

[0010] As described above, this invention utilizes random forests to evaluate feature importance, automatically selects the most effective feature combinations for distinguishing cells, constructs feature fingerprints that better characterize the uniqueness of cells, and thus forms multi-source data for identification and classification, further improving the adaptive capability and recognition accuracy of the identification method.

[0011] Optionally, step S1 further includes: Step S13: Using the cell sample library, generate a circular coverage area centered on the base station where sample broadband users in each cell most frequently stay at night, and perform a geographical union operation on the coverage areas of all base stations belonging to the same cell to form the digital boundary of each cell. Step S2 includes: Obtain the signaling data of the user to be identified, calculate the number of nighttime stay days of the user to be identified at each base station based on the signaling data of the user to be identified, and, based on the digital boundary of each cell, select all cells to which the base stations with the top O nighttime stay days fall as the candidate cell set, where O ≥ 2 and is a positive integer.

[0012] Optionally, step S1 further includes: Step S14: Construct a user-cell truth set using broadband account opening samples, and train a residential cell classification model based on the user-cell truth set.

[0013] Optionally, step S4 includes: Step S41: Obtain the family members of the user to be identified; Step S42: Aggregate the preliminary cell affiliation results of the user to be identified and the family members in the most recent M times, and select the cell to be identified as the final residential cell of the user to be identified. If there are multiple cells with the highest frequency, the most recently identified cell among the multiple cells with the highest frequency shall be the final residential cell of the user to be identified. M ≥ 2 and is a positive integer.

[0014] Optionally, the initial feature set includes residency features, time pattern features, population stability features, and resource coverage features.

[0015] Optionally, it also includes: Step S5: Periodically add newly registered broadband user data to the training set to update the residential community classification model, and use the updated residential community classification model to re-identify the residential communities of all users.

[0016] Optionally, the base station matching degree features include the coverage overlap between the top O base stations and the first candidate cell, and the total number of days the user to be identified stays at the base stations covered by the first candidate cell. The user behavior features are the dispersion of the user to be identified among the base stations active at night across the entire network and the number of times the International Mobile Equipment Identity (IMEI) is switched.

[0017] Optionally, the family member refers to all members in the same family group defined in the business system as the user to be identified.

[0018] In a second aspect, the present invention provides a residential community affiliation identification device based on multi-source data and machine learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the residential community affiliation identification method based on multi-source data and machine learning provided in the first aspect.

[0019] The technical effects of the residential community affiliation identification device based on multi-source data and machine learning provided in the second aspect are described in the relevant description of the residential community affiliation identification method based on multi-source data and machine learning provided in the first aspect. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the main process of the residential community affiliation identification method based on multi-source data and machine learning in an embodiment of the present invention; Figure 2 This is a schematic diagram of the residential community affiliation identification device based on multi-source data and machine learning according to an embodiment of the present invention.

[0021] Explanation of reference numerals in the attached figures: 1: A residential community affiliation identification device based on multi-source data and machine learning; 2: Processor; 3: Memory. Detailed Implementation

[0022] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0023] Example 1 Please refer to Figure 1 Residential community affiliation identification methods based on multi-source data and machine learning include: Step S1: Construct a community feature database using broadband account opening samples.

[0024] In this embodiment, step S1 includes: Step S11: Construct a community sample library using broadband account opening samples.

[0025] This involves extracting valid broadband user data from the operator's broadband service system for the past three months, including user ID and full address. Using an address resolution engine, the unstructured addresses are organized into five levels of structured addresses, forming a cell sample database. The structure of the cell sample information table stored in the cell sample database is shown in Table 1 below. Table 1. Community Sample Information Table

[0026] Step S12: Construct an initial feature set based on the signaling data of all broadband users in each cell in the cell sample library. Input the initial feature sets of all cells into the random forest model to evaluate the feature importance of each cell. Take the top N features with the highest importance scores in each cell as cell feature fingerprints. Construct a cell feature library from the cell feature fingerprints of all cells, where N≥2 and is a positive integer.

[0027] In step S11, each cell is used as a unit to aggregate the signaling data of all sample broadband users (no fewer than 20 households) to construct the initial feature set, which includes: (1) Core residency characteristics: STAY DAYS ALL (total number of days of stay per month), AVG STAY DURATION (average daily stay duration), NIGHT STAY RATE (percentage of nighttime stays).

[0028] (2) Time pattern characteristics: WORKDAY VS HOLIDAY RATIO (time spent on weekdays / holidays).

[0029] (3) Population stability characteristics: STABLE USER COUNT (number of users staying for more than 15 days per month) AVG IMEICHANGE FREQ (average IMEI change frequency per user).

[0030] (4) Resource coverage characteristics: CELL DENSITY (number of base stations per square kilometer), BROADBAND PENETRATION (number of broadband users / number of resident users).

[0031] Then, the random forest algorithm is used for feature selection: with the neighborhood category as the prediction target, the importance score I(f) of each feature is calculated. i For example, by calculating feature f i The importance of a feature is assessed by averaging the decrease in Gini impurity resulting from splitting all decision tree nodes. Finally, the top N features with the highest importance scores are selected to form the feature fingerprint vector Vcell=(v1,v2,...,v...). 15 Each feature value is normalized. The feature fingerprints of all cells constitute the cell feature database.

[0032] In this embodiment, N is 15. In other embodiments, N can be 3, 8, 12, 20, etc.

[0033] Step S13: Using the cell sample database, generate a circular coverage area centered on the base station where sample broadband users in each cell most frequently stay at night, and perform a geographical union operation on the coverage areas of all base stations belonging to the same cell to form the digital boundary of each cell.

[0034] In this embodiment, based on operator network planning data, the latitude and longitude (LAC, CELL ID, LON, LAT) of all base stations are obtained. The coverage radius is set according to the base station type, such as macro or micro, for example, 500 meters or 200 meters. A circular coverage area is generated centered on the base station where sample broadband users in each cell most frequently stay at night. The coverage areas of all base stations belonging to the same cell are then subjected to a geographic union operation to form an approximate digital boundary fence for that cell, replacing the need for crawling third-party map boundaries.

[0035] Step S14: Construct a user-community truth set using broadband account opening samples, and train a residential community classification model based on the user-community truth set.

[0036] In this embodiment, a user-cell truth set can be constructed using the data in Table 1 of the cell sample database to train a residential cell classification model. Specifically, the broadband sample user data is divided into a training set and a test set in a 7:3 ratio. The model's performance metrics, such as accuracy, recall, and F1 score, are evaluated on the test set.

[0037] Step S2: Obtain the signaling data of the user to be identified, and determine the candidate cell set based on the signaling data of the user to be identified.

[0038] In this embodiment, step S2 includes: Obtain the signaling data of the user to be identified, calculate the number of nighttime stay days of the user to be identified at each base station based on the signaling data of the user to be identified, and, based on the digital boundary of each cell, select all cells that the base stations with the highest number of nighttime stay days fall into as the candidate cell set, where O ≥ 2 and is a positive integer.

[0039] In this embodiment, signaling data of the user to be identified from 20:00 to 06:00 over the past 31 days is obtained, and its signaling data at each base station (LAC) is calculated. i CellId i ) number of days of stay D i Total duration T i Based on the digital cell boundaries generated in step S13, and sorted by the number of days spent in the cell, it is determined which cell boundaries the top five cell base stations where the user u is camped during the night fall into, forming a candidate cell set Candidates={c1,c2,...c j In this embodiment, Q is 5; in other embodiments, N can be 3, 6, 8, etc.

[0040] Step S3: Based on the user behavior characteristics obtained from the signaling data of the user to be identified, the cell feature fingerprint of the first candidate cell is obtained from the cell feature library. The user behavior characteristics, cell feature fingerprint, and base station matching degree characteristics between the user to be identified and the first candidate cell are input as input feature vectors into the pre-trained residential cell classification model to obtain the identification result of the first candidate cell. Then, the identification results of all candidate cells in the candidate cell set are obtained in turn to obtain the preliminary cell attribution result.

[0041] Among them, the base station matching degree features include: the top five base stations and the first candidate cell c j Coverage overlap R base =N common / 5, N common The first five base stations and the first candidate cell c j The coverage area; and the user to be identified in the first candidate cell c j Total number of days D of coverage base station damage cell .

[0042] Among them, user behavior characteristics include the dispersion of the user's activity base stations across the entire network at night and the number of times the International Mobile Equipment Identity (IMEI) is switched.

[0043] The residential community classification model is a multilayer perceptron (MLP) model, which outputs that the user u to be identified belongs to the first candidate community c. j The probability P(c) j |X u After completing the probability estimation of all candidate cells, preliminary cell assignment results can be obtained based on the preset probability threshold or the number of pre-screening and ranking results.

[0044] Step S4: Obtain the final residential community of the user to be identified based on the preliminary community affiliation results of the user's family members.

[0045] In this embodiment, step S4 includes: Step S41: Obtain the family members of the user to be identified.

[0046] In this context, family members refer to all members within the same family group defined in the business system as the user to be identified. These family groups include those defined in business systems such as family networks, family best-selling packages, and shared broadband guarantees.

[0047] Step S42: Aggregate the preliminary cell affiliation results of the user to be identified and their family members in the most recent M times, and select the cell to be identified as the final residential cell of the user to be identified. If there are multiple cells with the highest frequency, the most recently identified cell among the multiple cells with the highest frequency shall be the final residential cell of the user to be identified. M≥2 and is a positive integer.

[0048] In this embodiment, since the system updates and identifies cells regularly each month, a preliminary cell attribution result is generated each month. In this embodiment, M is 6, while in other embodiments it can be 3, 12, etc. That is, this embodiment aggregates the preliminary cell attribution results of the user and their family members from the past six months (the past 6 times), applying the principle of the highest frequency and the most recent time for aggregation. The most frequent principle: count all the neighborhoods that family members are identified in each month, and select the neighborhood with the highest frequency.

[0049] The latest time principle: If multiple cells have the same frequency, the cell identified in the most recent month will be selected.

[0050] Ultimately, the aggregated result was identified as the family's shared residential community, and this was used to correct any inconsistencies in the identification of individual members, significantly improving the accuracy of family-level identification.

[0051] Step S5: Periodically add newly registered broadband user data to the training set to update the residential community classification model, and use the updated residential community classification model to re-identify the residential communities of all users.

[0052] Therefore, all users are re-identified based on the updated model, enabling the model to adapt to network changes and user behavior migration.

[0053] In summary, by using random forest for feature importance assessment, the most effective feature combination for distinguishing cells is automatically selected, and a feature fingerprint that better represents the uniqueness of a cell is constructed. Then, the base station matching degree feature between the user to be identified and the first candidate cell, the user behavior feature of the user to be identified, and the cell feature fingerprint of the first candidate cell are used as multi-source data and input into the pre-trained residential cell classification model as input feature vectors, thereby significantly improving the adaptive ability and recognition accuracy of the identification method.

[0054] Example 2 Please refer to Figure 2 The residential community affiliation identification device 1 based on multi-source data and machine learning includes a memory 3, a processor 2, and a computer program stored on the memory 3 and run on the processor 2. When the processor 2 executes the computer program, it implements the steps in the above embodiment 1.

[0055] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0058] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0059] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0060] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0061] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for residential area attribution recognition based on multi-source data and machine learning, characterized in that, include: Step S1: Construct a community feature database using broadband account opening samples; Step S2: Obtain the signaling data of the user to be identified, and determine the candidate cell set based on the signaling data of the user to be identified; Step S3: Based on the user behavior features obtained from the signaling data of the user to be identified, and the cell feature fingerprint of the first candidate cell obtained from the cell feature library, the user behavior features, the cell feature fingerprint, and the base station matching degree features between the user to be identified and the first candidate cell are input as input feature vectors into the pre-trained residential cell classification model to obtain the identification result of the first candidate cell, and then obtain the identification results of all candidate cells in the candidate cell set in sequence to obtain the preliminary cell attribution result; Step S4: Obtain the final residential community of the user to be identified based on the preliminary community affiliation results of the family members of the user to be identified.

2. The residential community affiliation identification method based on multi-source data and machine learning according to claim 1, characterized in that, Step S1 includes: Step S11: Construct a community sample library using broadband account opening samples; Step S12: Construct an initial feature set based on the signaling data of all broadband users in each cell in the cell sample library. Input the initial feature sets of all cells into the random forest model to evaluate the feature importance of each cell. Take the top N features with the highest importance scores in each cell as cell feature fingerprints. Construct a cell feature library from the cell feature fingerprints of all cells, where N ≥ 2 and is a positive integer.

3. The residential community affiliation identification method based on multi-source data and machine learning according to claim 2, characterized in that, Step S1 further includes: Step S13: Using the cell sample library, generate a circular coverage area centered on the base station where sample broadband users in each cell most frequently stay at night, and perform a geographical union operation on the coverage areas of all base stations belonging to the same cell to form the digital boundary of each cell. Step S2 includes: Obtain the signaling data of the user to be identified, calculate the number of nighttime stay days of the user to be identified at each base station based on the signaling data of the user to be identified, and, based on the digital boundary of each cell, select all cells to which the base stations with the top O nighttime stay days fall as the candidate cell set, where O ≥ 2 and is a positive integer.

4. The residential community affiliation identification method based on multi-source data and machine learning according to claim 2, characterized in that, Step S1 further includes: Step S14: Construct a user-cell truth set using broadband account opening samples, and train a residential cell classification model based on the user-cell truth set.

5. The residential community affiliation identification method based on multi-source data and machine learning according to claim 2, characterized in that, The initial feature set includes residency features, time pattern features, population stability features, and resource coverage features.

6. The residential community affiliation identification method based on multi-source data and machine learning according to claim 1, characterized in that, Step S4 includes: Step S41: Obtain the family members of the user to be identified; Step S42: Aggregate the preliminary cell affiliation results of the user to be identified and the family members in the most recent M times, and select the cell to be identified as the final residential cell of the user to be identified. If there are multiple cells with the highest frequency, the most recently identified cell among the multiple cells with the highest frequency shall be the final residential cell of the user to be identified. M ≥ 2 and is a positive integer.

7. The residential community affiliation identification method based on multi-source data and machine learning according to any one of claims 1 to 5, characterized in that, The base station matching characteristics include the coverage overlap between the top O base stations and the first candidate cell, and the total number of days the user to be identified stayed at the base stations covered by the first candidate cell. The user behavior characteristics are the dispersion of the user to be identified among the base stations active at night across the entire network and the number of times the International Mobile Equipment Identity (IMEI) is switched.

8. The residential community affiliation identification method based on multi-source data and machine learning according to any one of claims 1 to 5, characterized in that, Also includes: Step S5: Periodically add newly registered broadband user data to the training set to update the residential community classification model, and use the updated residential community classification model to re-identify the residential communities of all users.

9. The residential community affiliation identification method based on multi-source data and machine learning according to any one of claims 1 to 5, characterized in that, The family members refer to all members in the same family group defined in the business system as the user to be identified.

10. A residential community affiliation identification device based on multi-source data and machine learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the residential community affiliation identification method based on multi-source data and machine learning as described in any one of claims 1 to 9.

Images