Travel user identification method and system based on communication data

Abstract

The invention discloses a tourist user identification method and system based on communication data, and relates to the technical field of communication algorithms, and the method comprises the specific steps: crawling each scenic spot list, selecting a plurality of data sources for crawling and data fusion, and obtaining related data of a target scenic spot, including scenic spot names, scenic spot scores and scenic spot comments, the scenic spot data are crawled from a plurality of tourism websites through a multi-source data fusion technology, rapid acquisition and fusion of the scenic spot data are realized, in the data processing process, travel sequence data of the users are automatically constructed by utilizing an advanced data processing and analysis technology, and through processing of user communication data, the scenic spot comment number is collected, so that the user experience is improved. The system can automatically identify the resident residence of the user, generate a detailed travel track, and perform screening and feature association on the travel points, so that the generated travel table can more comprehensively reflect the actual behavior of the user.

Description

Technical Field

[0001] The present invention relates to the technical field of communication algorithms, and in particular to a method and system for identifying tourist users based on communication data. Background Art

[0002] With the rapid development of tourism and the popularization of mobile Internet, tourist management has become a key link in improving tourism experience and scenic spot operation efficiency. Traditional tourist identification methods mostly rely on manual statistics and questionnaires, which are not only inefficient, but also difficult to fully and accurately reflect the actual behavior and needs of tourists. Therefore, it is particularly important to explore automated tourist identification methods based on big data and communication technology.

[0003] Although communication technology makes it possible to identify tourists, traditional location-based services (LBS) mainly rely on GPS data, which poses a risk of privacy leakage and is difficult to obtain data. On the other hand, the identification method based on base station signaling data has relatively low accuracy and is difficult to meet the needs of refined management. In addition, the existing technology lacks a technical solution that can integrate multiple communication data to achieve efficient and accurate tourist identification.

[0004] Therefore, the present invention proposes a tourist user identification method and system based on communication data, which provides new ideas and solutions for the intelligent management and personalized services of the tourism industry. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and to provide a tourist user identification method and system based on communication data. It can utilize the user's O-domain data, B-domain data and relevant data of tourist attractions, and through advanced data processing and analysis technology, construct the user's itinerary trajectory, match the user's itinerary trajectory with the location of tourist attractions, and realize comprehensive and accurate identification of tourist users.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: On the one hand, a tourist user identification method based on communication data, the specific steps of the method are:

[0007] S1. Acquisition of scenic spot data: crawling the data of each scenic spot, that is, selecting multiple data sources for crawling and data fusion, obtaining the relevant data of the target scenic spot, including the name of the scenic spot, the score of the scenic spot, the comments of the scenic spot, and summarizing the number of comments on the scenic spot;

[0008] Based on the collected data, the scenic area attributes of scenic spots and scenic spots are obtained and associated, and the scenic area attributes include whether it is a 4A scenic area, whether it is a 3A scenic area, longitude and latitude, and area of ​​the scenic area;

[0009] S2. Construct user itinerary sequence data: For the same user, the places passed by in a continuous period of time are a data set consisting of multiple data points, which is the itinerary data of the user: <the starting and ending points are the permanent residence, and the places passed on the way are the content of the itinerary>. The permanent residence introduces a screening logic to filter out the permanent residence that has not appeared for 4 consecutive months;

[0010] Eliminate the trip point data with a stay time of less than half an hour in the trip set;

[0011] Use the data in the filtered and eliminated user trip set to associate user features and generate a trip point vertical table, where trace_id is a trip ID used to identify the trip category of the trip point, and the trace_id generation rule is: user number + timestamp corresponding to the trip start point, that is, trace_id = user_number + start_time_stamp, where user_number represents the user number, and start_time_stamp represents the timestamp corresponding to the trip start point;

[0012] Convert the vertical table of itineraries into a horizontal table, and fix the number of points for each itinerary in the itinerary table to 10. That is, for itineraries with less than 10 points, fill in the missing points with blank values. For itineraries with more than 10 points, keep the first 10 points. The number of fields in the horizontal table is the number of fields * 10.

[0013] The user trips are organized into sequences, that is, the features are converted from rows to columns. The dimension of the constructed features is (n_data, n_xc, n_feas), where n_data is the number of sequence data, n_xc is the trip sequence length, and n_feas is the number of feature columns.

[0014] S3, constructing travel itinerary identification rules: determining the rule judgment condition factors of tourists, the judgment condition factors include whether it is a day trip, whether it is an overnight trip, whether it is a hotel overnight, the distance from the permanent residence, the length of hotel stay, and the itinerary points of the tourist target;

[0015] The analytic hierarchy process is used to construct a travel itinerary scoring matrix, and the confidence value of the itinerary belonging to the travel itinerary is calculated based on the determined weights of the influencing factors;

[0016] Perform weighted scoring according to the weights output by the hierarchical analysis method, output the final score value, and obtain the rule prediction result that each user is a travel customer;

[0017] S4, training a travel itinerary recognition model: removing the itinerary corresponding to the user's permanent residence from the basic data set, where the itinerary corresponding to the permanent residence includes the permanent workplace, school, and park, and associating the park, workplace, and school through the POI dimension table;

[0018] Use manual labeling to annotate some data sets and mark whether the itinerary belongs to a travel itinerary;

[0019] Performing feature engineering processing on the annotated travel data, the feature engineering processing includes: processing outliers and null values ​​on the data set, digitizing map POI features, and standardizing the entire data set;

[0020] Use the labeled sequence data set to train the model, predict the user's itinerary sequence data, and obtain the probability that each user is a travel customer;

[0021] S5. Output the travel itinerary prediction results and identify the travel customers: weight the prediction probability output by the model and the prediction result output by the rule, and set the model weight to 60% and the rule weight to 40%;

[0022] Label the user's itinerary sequence data, including:

[0023] If the itinerary includes scenic spots, it will be marked as "scenic spot tour itinerary";

[0024] For the "attraction tour itinerary", further annotation is performed based on the attractions in the itinerary, multiple attractions are separated by "|", and the annotated "attraction tour itinerary" data is output, including user ID, itinerary start time, itinerary end time, and attraction annotation information;

[0025] According to the weighted prediction score and itinerary marking results, the identification criteria for tourist customers are formulated and the threshold is set. That is, when the weighted prediction score is greater than the threshold and the itinerary is marked as "attraction tour itinerary" and "multi-attraction tour itinerary", it is identified as a tourist customer, and a list of identified tourist customers is output, including user ID, itinerary information and prediction score information.

[0026] Furthermore, when S2 converts the vertical table of the trip into a horizontal table during the process of constructing the user trip sequence data, each trip point in the vertical table of the trip contains p fields, and the fields of each trip point in the horizontal table are numbered f1, f2, ..., f 10p For a trip with less than 10 trip points, the rule for filling in blank values ​​is: when the trip contains q trip points and q<10, fill in blank values ​​starting from the (q+1)th trip point, that is, the field f corresponding to the (q+1)th trip point (q×p+1) to f ((q+1)×p) Fill in empty values ​​until the field f corresponding to the 10th trip point (9×p+1) to f 10p For itineraries with more than 10 travel points, the first 10 travel points are retained by directly selecting the data of the first 10 travel points and filling them into the corresponding field positions in the horizontal table.

[0027] Furthermore, when constructing the user trip generation sequence data, S2 considers that there are m trips in the user trip set, and the i-th trip contains n i There are n travel points and i ≤10, then the number of sequence data is n data =m, the length of the run sequence is n xc =10, number of feature columns n feas =p×10, p is the number of fields contained in each trip point in the trip table. The sequence data is generated by arranging the eigenvalues ​​of each trip point in order to form a eigenvector. The eigenvectors of m trips are combined into sequence data, that is, for the i-th trip, its j-th trip point j≤n i The eigenvalues ​​of are filled into the positions (i-1)×10×p+(j-1)×p+1 to (i-1)×10×p+j×p in the sequence data in sequence. i , the corresponding position is filled with null value.

[0028] Furthermore, in the S3, the analytic hierarchy process is used to construct a travel itinerary scoring matrix in the travel itinerary identification rule, and the formula for constructing the travel itinerary scoring matrix is: where a ij Indicates the relative importance of the i-th judgment condition factor relative to the j-th judgment condition factor and satisfies a ii =1, n is the number of influencing factors, including whether it is a one-day tour, whether it is an overnight tour, whether the hotel is overnight and the distance from the permanent residence, the length of hotel stay, and the itinerary points of the tourist target.

[0029] Furthermore, S3 calculates the confidence value that the itinerary belongs to the travel itinerary based on the determined weight of the influencing factor, and the confidence value C that the itinerary belongs to the travel itinerary is: where w i is the weight of the i-th judgment condition factor, f i is the eigenvalue of the i-th judgment condition factor, and n is the number of judgment condition factors.

[0030] Furthermore, in the step of constructing the travel itinerary identification rule, the influencing factor weight w i The formula is determined as: where w i represents the weight of the i-th judgment condition factor, a i It represents the importance value of the i-th judgment condition factor in the travel itinerary scoring matrix M, that is, the sum of the i-th row elements in the matrix M, n is the number of judgment condition factors, and according to the constructed travel itinerary scoring matrix M, it is calculated Get the weight w i .

[0031] Furthermore, in the process of outputting the travel itinerary prediction results and identifying the travel customers, the weighted score S of S5 is: S = 0.6 × P m +0.4×P r , where P m Represents the predicted probability of the model output, P r Represents the prediction result of the rule output, the prediction probability P of the model output m After normalization, the predicted result P of the rule output is between 0 and 1. r After quantization, it is also between 0 and 1. The specific quantification method is: according to the confidence value C obtained in the step of building the travel itinerary identification rule, set the threshold T1. When C ≥ T1, P r =1, otherwise P r =0.

[0032] On the other hand, a tourist user identification system based on communication data, the system comprises: a scenic spot data collection module, a trip sequence construction module, a rule construction module, a trip sequence prediction module, and an identification output module;

[0033] The scenic spot data collection module is responsible for crawling and integrating scenic spot data from various data sources to obtain scenic spot names, rating information and related attributes, including scenic spot logos, longitude and latitude;

[0034] The itinerary sequence construction module determines the starting and ending points of the itinerary as the permanent residence based on the user communication data, processes the itinerary point data, associates the user characteristics to generate a vertical table, converts the horizontal table and constructs sequence data to characterize the user's itinerary trajectory;

[0035] The rule building module clarifies the tourist rule judgment factors, builds a scoring matrix through the hierarchical analysis method to determine the weights, calculates the confidence value and weighted scores, and provides a basis for the judgment of the travel itinerary;

[0036] The trip sequence prediction module removes the trips corresponding to the permanent place, manually annotates part of the data set, performs feature engineering on the annotated data, including outlier and null value processing, POI feature digitization and standardization, and uses the sequence data set to train the model;

[0037] The identification output module weights the output results of the model and the rules, annotates the user's itinerary data, sets a threshold according to the weighted score and itinerary annotation to identify the tourist customers, and outputs a customer list.

[0038] Compared with the prior art, the tourist user identification method and system based on communication data has the following beneficial effects:

[0039] 1. The present invention crawls scenic spot data from multiple travel websites through multi-source data fusion technology, thereby realizing the rapid acquisition and fusion of scenic spot data. In the data processing process, the present invention utilizes advanced data processing and analysis technology to automatically construct the user's itinerary sequence data. By processing the user's communication data, the system can automatically identify the user's permanent residence, generate a detailed itinerary trajectory, and screen and feature-associate itinerary points. The generated itinerary table can more comprehensively reflect the user's actual behavior, providing a solid foundation for subsequent travel itinerary identification.

[0040] 2. The tourist user identification method based on communication data of the present invention, on the one hand, by accurately identifying tourist users, tourism enterprises can push personalized tourism products and services more accurately, improve customer satisfaction and loyalty, and thus promote the growth of tourism business. On the other hand, the application of this method also contributes to the rational allocation and optimal utilization of tourism resources, reveals the preferences and behavioral patterns of tourist users through data analysis, monitors and analyzes tourist user data, provides data reference for the formulation and adjustment of tourism policies, and promotes the healthy and orderly development of the tourism industry.

[0041] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 It is an operation flow chart of a tourist user identification method based on communication data;

[0044] Figure 2 This is a module composition diagram of a tourist user identification system based on communication data. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Embodiment 1

[0047] This embodiment describes in detail the complete process of a tourist user identification method based on communication data. It obtains scenic spot data through multiple data sources, uses user communication data to build an itinerary sequence, and combines tourist itinerary identification rules and models to achieve accurate identification of tourist users, providing strong support for tourism industry management and services.

[0048] First, this embodiment selects Ctrip.com, Meituan.com and Qyer.com as the main data sources to obtain scenic spot data. For each website, a special crawling script is formulated to ensure that the relevant information of the scenic spots in the target province can be accurately captured. The crawled information includes the scenic spot name, scenic spot score, scenic spot comments and various attributes of the scenic spot, such as scenic spot grade, longitude and latitude, and scenic spot area. By analyzing its page structure and data storage method, the elements of the scenic spot list page are located using web crawler technology, and the name, score, number of comments and other information of the scenic spot are extracted. The data obtained from different data sources are fused, and the scenic spot name is used as the key identifier to match the data from the three data sources. If the names are exactly the same, they are considered to be data from the same scenic spot. In the fusion process, the scenic spot attributes, namely the scenic spot grade, longitude and latitude, and scenic spot area, are fused first. The supplementary information of other data sources is recorded in additional fields to form a complete scenic spot data record. At the same time, the number of scenic spot comments is summarized. By accumulating the number of comments on the scenic spot in each data source, the total number of comments on the scenic spot is obtained, which provides data support for the subsequent analysis of the popularity and popularity of the scenic spot. The user's communication data is collected to obtain the user's location change data in continuous time. These data are recorded in the form of time series. Each record contains user number, timestamp, location coordinates and other information. According to the defined itinerary data rules, the original communication data is filtered to filter out the location data passed by the same user in continuous time, and the starting and ending points are required to be the permanent residence. In the preprocessing process, the data is cleaned to remove erroneous or invalid data records, and the timestamps are uniformly converted into time format.

[0049] Then, the permanent residence screening logic is introduced to analyze the user's location data, and the number of days the user appears in each place in 4 consecutive months is counted. If the number of days the user appears in each month in these 4 months is less than the day threshold, the place is determined to be a permanent place that has not appeared in 4 consecutive months, and it is excluded from the start and end point consideration range of the itinerary data. The itinerary point data with a stay time of less than half an hour in the itinerary set are eliminated. By calculating the user's stay time at each itinerary point (determined according to the time interval of continuous location data), the itinerary points with a stay time of less than half an hour are deleted to exclude the user's short stay at non-tourism related places. For the remaining itinerary point data, these data are used to associate user features to generate a itinerary point vertical table, where trace_id is the itinerary ID, which is generated according to the rule of user number + timestamp corresponding to the itinerary start point. The itinerary vertical table is converted into a horizontal table. Each itinerary point in the itinerary vertical table is assumed to contain p fields (location information, arrival time, departure time, etc.), and the fields of each itinerary point in the horizontal table are numbered f1, f2, ..., f 10p For a trip with less than 10 trip points, if the trip contains q trip points (q<10), the null value is filled starting from the (q+1)th trip point, that is, the field f corresponding to the (q+1)th trip point (q×p+1) to f ((q+1)×p) Fill in empty values, and so on until the field f corresponding to the 10th trip point (9×p+1) to f 10p For trips with more than 10 trip points, directly select the data of the first 10 trip points and fill them into the corresponding field positions in the horizontal table. The user trips are organized into a sequence. Suppose there are m trips in the user trip set, and the i-th trip contains n i Travel points (n i ≤10), then the number of sequence data is n data =m, the length of the run sequence is n xc =10, number of feature columns n f eas=p×10, the sequence data is generated by arranging the eigenvalues ​​of each trip point in order to form a eigenvector, and the eigenvectors of m trips are combined into sequence data, that is, for the i-th trip, its j-th trip point (j≤n i ) are filled into the positions (i-1)×10×p+(j-1)×p+1 to (i-1)×10×p+j×p in the sequence data in sequence. If j>n i , the corresponding position is filled with null value.

[0050] Subsequently, the rule judgment condition factors of tourists are determined, including the type of travel itinerary (divided into one-day tour and overnight tour, and overnight tour is further divided into hotel overnight and self-stayed overnight), distance from the permanent residence, hotel stay time, and itinerary points of tourist objects. The distance from the permanent residence is determined by calculating the actual geographical distance between the user's itinerary point and the permanent residence (the permanent residence after screening). The hotel stay time is calculated according to the user's stay time in the hotel-related area (determined by associating with the hotel POI data). The itinerary points of tourist objects are calculated by counting the number of tourist attractions passed by the user in the itinerary (determined according to the latitude and longitude range in the scenic spot data). The hierarchical analysis method is used to construct the travel itinerary scoring matrix M, and four judgment condition factors are defined (travel itinerary type, distance from the permanent residence, hotel stay time, and itinerary points of tourist objects). Then the matrix Matrix element a ij Indicates the relative importance of the i-th judgment condition factor relative to the j-th judgment condition factor, with a value range of 1-9 and meeting a ii =1, after constructing the matrix M, calculate the weight vector W of each judgment condition factor, the formula is After calculating the weight vector W, the weight w of each judgment condition factor is obtained i , based on the determined weights of the influencing factors, the confidence value C of the itinerary belonging to the travel itinerary is calculated, and the formula is C = ∑ i＝1 w i ×f i , where f i is the characteristic value of the th judgment condition factor, and a weighted score is performed according to the weight output by the hierarchical analysis method. The final score value is output to obtain the rule prediction result that each user is a tourist customer.

[0051] Next, the itinerary corresponding to the user's permanent residence is removed from the basic data set. The parks, workplaces, schools and other permanent residence-related places are associated through the POI dimension table. The itinerary data related to these places are removed. Some data sets are annotated manually, and the annotated itineraries are matched with travel itineraries. According to the definition and characteristics of travel itineraries, the data sets are annotated as travel itineraries or non-travel itineraries. During the annotation process, feature engineering is performed on the annotated itinerary data, that is, outliers and null values ​​are processed on the data set. The outliers in the data set are identified through statistical analysis methods, and they are corrected and deleted according to the distribution of the data. Different types of POIs (are converted into digital codes so that the model can process them. The annotated sequence data set is used for model training. The model LSTM that supports sequence data input is selected. The processed data set is divided into training set, validation set and test set according to the proportion. During the training process, the model parameters are set, and the loss function and accuracy of the model on the training set and validation set are observed. The parameters are continuously adjusted until the model performance reaches the best.

[0052] Finally, the predicted probability P of the model output m And the prediction result P output by the rule r Weighted, weighted score S calculation formula is S = 0.6 × P m +0.4×P r , the predicted probability P of the model output m Normalize it to be between 0 and 1, and the prediction result P output by the rule r After quantization, it is also between 0 and 1. The specific quantification method is: according to the confidence value C obtained in the step of building the travel itinerary identification rule, set the threshold T1. When C ≥ T1, P r =1, otherwise P r =0, and the user's itinerary sequence data is annotated. If the itinerary includes a scenic spot, it is annotated as a "scenic spot tour itinerary". For the "scenic spot tour itinerary", it is further annotated based on the most point in the itinerary. Multiple scenic spots are separated by "|", and the start time and end time of the itinerary are recorded at the same time. According to the weighted prediction score and itinerary annotation results, the identification standard of tourist customers is formulated and the threshold is set. When the weighted prediction score is greater than the threshold and the itinerary is annotated as a "scenic spot tour itinerary" or "multi-scenic spot tour itinerary", it is identified as a tourist customer, and the identified tourist customer list is output, including user ID, itinerary information (such as itinerary start time, end time, scenic spots passed, etc.) and prediction score. This information can provide accurate customer source information for tourism enterprises, help enterprises formulate personalized tourism product recommendations and marketing strategies, and also provide data support for tourism management departments for the reasonable allocation of tourism resources and management decisions.

[0053] In summary, this embodiment demonstrates in detail the implementation process of the tourist user identification method based on communication data, from the scenic spot data acquisition and user communication data collection in the data preparation stage, to the itinerary sequence construction, identification rule construction and model training in the user itinerary analysis stage, and then to the prediction result weighting and final identification in the tourist user identification stage. Through multi-source data fusion, advanced data processing technology and reasonable model construction and application, accurate identification of tourist users is achieved, which provides strong technical support for the development of the tourism industry, helps to improve the service quality and market competitiveness of tourism enterprises, and is also conducive to the optimal allocation of tourism resources and the sustainable development of the industry.

[0054] Embodiment 2

[0055] This embodiment focuses on the construction and application of travel itinerary identification rules in the travel user identification method based on communication data. It elaborates on how to determine the rule judgment condition factors, use the hierarchical analysis method to construct a scoring matrix, calculate the confidence value and perform weighted scoring, and finally achieve effective identification of travel itineraries, providing key decision-making basis for the entire travel user identification system.

[0056] First, the rule judgment condition factors of tourists are clarified, including the type of travel itinerary, the distance from the permanent residence, the length of hotel stay, and the number of itinerary points of tourist objects. The travel itinerary type is divided into one-day tour and overnight tour, and the overnight tour is further divided into hotel overnight and self-stay overnight. The distance from the permanent residence is calculated by calculating the actual geographical distance between the user's itinerary point and the permanent residence (the permanent residence after screening). The length of hotel stay is determined according to the user's stay time in the hotel-related area (accurately associated with the hotel POI data). The number of itinerary points of tourist objects is based on the longitude and latitude range in the scenic area data, and the number of tourist attractions passed by the user in the itinerary is counted. The location information is extracted from the user's communication data, and all the data required for the rule judgment factors are collected in combination with the scenic area POI data and the hotel POI data. For the location information, its accuracy and completeness are ensured. The wrong or invalid location data points are removed through the data cleaning algorithm. The hotel POI data is updated and integrated to ensure accurate matching with the hotel information where the user actually stays. The longitude and latitude range data of the tourist attractions are sorted to make it consistent with the format and coordinate system of the user's itinerary data, so as to make accurate itinerary point statistics.

[0057] Then, the hierarchical analysis method is used to construct the tourism itinerary scoring matrix M, and the judgment condition factors are defined as 4 (tourism itinerary type, distance from the permanent residence, hotel stay time, and tourism target itinerary points). Then the matrix M is a 4×4 square matrix, and the matrix element a ij Indicates the relative importance of the i-th judgment condition factor relative to the j-th judgment condition factor, with a value range of 1-9 and meeting a ii =1, determine the relative importance of each factor, and calculate the weight vector W of each judgment condition factor. The formula is: For each row i of the matrix M, calculate the ratio of the row element to the sum of all column elements, that is, Then, these ratios are averaged in the column direction to obtain the i-th element w of the weight vector W i , that is, for the first row, calculate ( Get w1, and then calculate all the elements of the weight vector W, so as to get the weight w of each judgment condition factor i .

[0058] The confidence value C of the itinerary belonging to the travel itinerary is calculated based on the determined weight of the influencing factors, and the formula is C = ∑ i-1 w i ×f i , where f i is the eigenvalue of the th judgment condition factor. During the calculation process, ensure that the eigenvalue f i The accuracy and rationality of the rule prediction are as follows. For characteristic values ​​such as distance and duration, the collected and calculated data are directly used. For classification characteristic values ​​such as type, they are converted according to quantitative rules, and weighted scores are performed according to the weights output by the hierarchical analysis method. The final score value is output to obtain the rule prediction result that each user is a tourist customer. The confidence value C is compared with the set threshold. If C ≥ the threshold, the trip is judged to be a tourist trip, and the rule prediction result is 1; otherwise, it is judged to be a non-tourist trip, and the rule prediction result is 0. At the same time, the detailed information of each trip is recorded, including user ID, trip start time, trip end time, places passed by the trip, etc., as well as the corresponding confidence value and rule prediction results, to provide complete data support for subsequent comprehensive analysis and tourism customer identification.

[0059] Finally, an integrated interface between the rules and the entire tourism user identification system is designed to ensure that the rules can be seamlessly integrated into the system's data flow. The interface defines the data input format and receives travel data from the user itinerary sequence construction module, including itinerary point information, user characteristics, etc. At the same time, the data output format is defined to pass the confidence value, prediction results, and related itinerary annotation information calculated by the rules to subsequent modules. During the operation of the system, when new user itinerary data enters the system, the rule construction module immediately processes it, calculates the confidence value and prediction results, and feeds the results back to the system. The system makes a comprehensive decision based on the output results of the rules and the itinerary sequence prediction module. At the same time, a feedback mechanism is established to collect the actual effect data after the rule application, adjust the weight value in the weight vector W according to the actual situation, or modify the quantization method of the judgment condition factor, so that the rules can better adapt to different tourism scenarios and user behavior patterns, and continuously improve the accuracy and effectiveness of tourism itinerary identification.

[0060] In summary, this embodiment demonstrates in detail the entire process of constructing and applying travel itinerary identification rules based on communication data, from the determination and quantification of rule judgment factors, to the construction of a travel itinerary scoring matrix and weight calculation, to confidence value calculation, weighted scoring, and integration and application of rules in the system. Through precise data processing, reasonable algorithm application, and effective system integration, it provides an important decision-making basis for the travel user identification system, which helps to improve the accuracy and scientificity of tourism industry management and services.

[0061] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A tourist user identification method based on communication data, characterized in that: The specific steps of this method are: S1. Acquisition of scenic spot data: crawling the data of each scenic spot, that is, selecting multiple data sources for crawling and data fusion, obtaining the relevant data of the target scenic spot, including the name of the scenic spot, the score of the scenic spot, the comments of the scenic spot, and summarizing the number of comments on the scenic spot; Based on the collected data, the scenic area attributes of the scenic area and scenic spot are obtained and associated, and the scenic area attributes include the scenic area grade, longitude and latitude, and scenic area; S2. Construct user itinerary sequence data: For the same user, the places passed by in a continuous period of time are a data set consisting of multiple data points, which is the itinerary data of the user: <the starting and ending points are the permanent residence, and the places passed on the way are the content of the itinerary>. The permanent residence introduces a screening logic to filter out the permanent residence that has not appeared for 4 consecutive months; Eliminate the trip point data with a stay time of less than half an hour in the trip set; Use the data in the filtered and eliminated user itinerary set to associate user features and generate a vertical table of itinerary points, where trace_id is a itinerary ID used to identify the itinerary category of the itinerary point, and the trace_id generation rule is: user number + timestamp corresponding to the starting point of the itinerary, that is, trace_id = user_number + start_time_stamp, where user_number represents the user number, and start_time_stamp represents the timestamp corresponding to the starting point of the itinerary; Convert the vertical table of itineraries into a horizontal table, and fix the number of points for each itinerary in the itinerary table to 10. That is, for itineraries with less than 10 points, fill in the missing points with blank values. For itineraries with more than 10 points, keep the first 10 points. The number of fields in the horizontal table is the number of fields * 10. The user trips are organized into sequences, that is, the features are converted from rows to columns. The dimension of the constructed features is (n_data, n_xc, n_feas), where n_data is the number of sequence data, n_xc is the trip sequence length, and n_feas is the number of feature columns. S3, constructing travel itinerary identification rules: determining the rule judgment condition factors of tourists, wherein the judgment condition factors include travel itinerary type, distance from the permanent residence, hotel stay duration, and travel points of the tourist target. The travel itinerary type is divided into one-day tour and overnight tour, wherein the overnight tour is further divided into hotel overnight and self-stayed overnight; The analytic hierarchy process is used to construct a travel itinerary scoring matrix, and the confidence value of the itinerary belonging to the travel itinerary is calculated based on the determined weights of the influencing factors; Perform weighted scoring according to the weights output by the hierarchical analysis method, output the final score value, and obtain the rule prediction result that each user is a travel customer; S4, training a travel itinerary recognition model: removing the itinerary corresponding to the user's permanent residence from the basic data set, where the itinerary corresponding to the permanent residence includes the permanent workplace, school, and park, and associating the park, workplace, and school through the POI dimension table; Manual annotation was used to annotate some data sets and the annotated itineraries were matched with travel itineraries; Performing feature engineering processing on the annotated travel data, the feature engineering processing includes: processing outliers and null values ​​on the data set, digitizing map POI features, and standardizing the entire data set; Use the labeled sequence data set to train the model, predict the user's itinerary sequence data, and obtain the probability that each user is a travel customer; S5. Output the travel itinerary prediction results and identify the travel customers: weight the prediction probability output by the model and the prediction result output by the rule, and set the model weight to 60% and the rule weight to 40%; Label the user's itinerary sequence data, including: If the itinerary includes scenic spots, it will be marked as "scenic spot tour itinerary"; For the "attraction tour itinerary", further annotation is performed based on the attractions in the itinerary, multiple attractions are separated by "|", and the annotated "attraction tour itinerary" data is output, including user ID, itinerary start time, itinerary end time, and attraction annotation information; According to the weighted prediction score and itinerary marking results, the identification standard of tourist customers is formulated and the threshold is set. That is, when the weighted prediction score is greater than the threshold and the itinerary is marked as "attraction tour itinerary" and "multi-attraction tour itinerary", it is identified as a tourist customer, and a list of identified tourist customers is output, including user ID, itinerary information and prediction score information.

2. A tourist user identification method based on communication data according to claim 1, characterized in that: When S2 converts the vertical table of the trip into a horizontal table in the process of constructing the user trip sequence data, each trip point in the vertical table contains p fields, and the fields of each trip point in the horizontal table are numbered f1, f2, ..., f 10p For a trip with less than 10 trip points, the rule for filling in blank values ​​is: when the trip contains q trip points and q<10, fill in blank values ​​starting from the (q+1)th trip point, that is, the field f corresponding to the (q+1)th trip point (q×p+1) to f ((q+1)×p) Fill in empty values ​​until the field f corresponding to the 10th trip point (9×p+1) to f 10p For itineraries with more than 10 travel points, the first 10 travel points are retained by directly selecting the data of the first 10 travel points and filling them into the corresponding field positions in the horizontal table.

3. A tourist user identification method based on communication data according to claim 1, characterized in that: When constructing the user trip generation sequence data, S2 has a total of m trips in the user trip set, and the i-th trip contains n i There are n travel points and i ≤10, then the number of sequence data is n data =m, the length of the run sequence is n xc =10, number of feature columns n feas =p×10, p is the number of fields contained in each trip point in the trip table. The sequence data is generated by arranging the eigenvalues ​​of each trip point in order to form a eigenvector. The eigenvectors of m trips are combined into sequence data, that is, for the i-th trip, its j-th trip point j≤n i The eigenvalues ​​of are filled into the positions (i-1)×10×p+(j-1)×p+1 to (i-1)×10×p+j×p in the sequence data in sequence. i , the corresponding position is filled with null value.

4. A tourist user identification method based on communication data according to claim 1, characterized in that: The S3 constructs the travel itinerary identification rules and uses the hierarchical analysis method to construct the travel itinerary scoring matrix, and the travel itinerary scoring matrix construction formula is: where a ij Indicates the relative importance of the i-th judgment condition factor relative to the j-th judgment condition factor and satisfies a ii =1, n is the number of influencing factors, including travel itinerary type, distance from the permanent residence, hotel stay time, and travel target itinerary points.

5. A tourist user identification method based on communication data according to claim 1, characterized in that: The S3 calculates the confidence value of the itinerary belonging to the travel itinerary based on the determined weight of the influencing factors, and the confidence value C of the itinerary belonging to the travel itinerary is: where w i is the weight of the i-th judgment condition factor, f i is the eigenvalue of the i-th judgment condition factor, and n is the number of judgment condition factors.

6. A tourist user identification method based on communication data according to claim 1, characterized in that: In the step of constructing travel itinerary identification rules, the influencing factor weight w i The formula is: where w i represents the weight of the i-th judgment condition factor, a i It represents the importance value of the i-th judgment condition factor in the travel itinerary scoring matrix M, that is, the sum of the i-th row elements in the matrix M, n is the number of judgment condition factors, and according to the constructed travel itinerary scoring matrix M, it is calculated Get the weight w i .

7. A tourist user identification method based on communication data according to claim 1, characterized in that: In the process of outputting the prediction results of the travel itinerary and identifying the travel customers, the weighted score S of S5 is: S = 0.6 × P m +0.4×P r , where P m Represents the predicted probability of the model output, P r Represents the prediction result of the rule output, the prediction probability P of the model output m After normalization, the predicted result P of the rule output is between 0 and 1. r After quantization, it is also between 0 and 1. The specific quantification method is: according to the confidence value C obtained in the step of building the travel itinerary identification rule, set the threshold T1. When C ≥ T1, P r =1, otherwise P r =0.

8. A tourist user identification system based on communication data, characterized in that: The system consists of: scenic spot data collection module, itinerary sequence construction module, rule construction module, itinerary sequence prediction module, and recognition output module; The scenic spot data collection module is responsible for crawling and integrating scenic spot data from various data sources to obtain scenic spot names, rating information and related attributes, including scenic spot logos, longitude and latitude; The itinerary sequence construction module determines the starting and ending points of the itinerary as the permanent residence based on the user communication data, processes the itinerary point data, associates the user characteristics to generate a vertical table, converts the horizontal table and constructs sequence data to characterize the user's itinerary trajectory; The rule building module clarifies the tourist rule judgment factors, builds a scoring matrix through the hierarchical analysis method to determine the weights, calculates the confidence value and weighted scores, and provides a basis for the judgment of the travel itinerary; The trip sequence prediction module removes the trips corresponding to the permanent place, manually annotates part of the data set, performs feature engineering on the annotated data, including outlier and null value processing, POI feature digitization and standardization, and uses the sequence data set to train the model; The identification output module weights the output results of the model and the rules, annotates the user's itinerary data, sets a threshold according to the weighted score and itinerary annotation to identify the tourist customers, and outputs a customer list.

Images