A personalized travel route recommendation method based on feature mining algorithm

Through feature mining algorithms and deep learning technology, combined with user habits and environmental information, personalized travel route recommendations and push return routes are provided, which solves the problem that traditional platforms cannot meet users' personalized needs and improves the personalization and satisfaction of recommendations.

CN114020993BActive Publication Date: 2024-11-01XIAN TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110584507.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-05-27
Publication Date
2024-11-01
Estimated Expiration
2041-05-27

AI Technical Summary

Technical Problem

Traditional travel platforms cannot provide targeted and personalized route recommendations, lack return route planning functions, and cannot meet users' travel needs to meet their personal preferences.

Method used

Adopt personalized travel route recommendation method based on feature mining algorithm, collect basic user information, establish user feature models, use deep learning technology and data analysis technology to mine users' travel habits and preferences, recommend personalized routes, and use correlation Analysis provides return routes.

Benefits of technology

It improves the personalization and satisfaction of the recommendation results, can record the user's travel route for multiple use, and push routes that meet user needs when returning, improving the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114020993B_ABST
    Figure CN114020993B_ABST
Patent Text Reader

Abstract

The present invention relates to a personalized travel route recommendation method based on a feature mining algorithm, comprising the following steps: step 1, collecting basic information of users and establishing a user data information database; step 2, analyzing the user information and establishing a user feature model; step 3, recommending personalized travel routes for users. The method combines data analysis technology and deep learning technology, explores user interest features as much as possible based on a data mining algorithm, improves the personalization and satisfaction of recommendation results, adds the traveled routes to the search results, and then conducts deep learning through user feedback, thereby obtaining a path planning result that better meets user needs. At the same time, it can not only record and save the travel routes that users have traveled, so that the users can directly use them when they travel next time, but also directly push out the return route according to the analysis of the personalized outbound route finally recommended by the system when preparing for the return trip, so that the users can directly use it when they return.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of personalized route setting, and in particular relates to a personalized travel route recommendation method based on a feature mining algorithm. Background Art

[0002] With the emergence of mobile Internet maps, people's travel has been greatly facilitated. However, in the face of the massive growth of map data, excessive information has become the most direct problem, and the need to improve the personalized experience of users has become very urgent. Traditional travel platforms have too many recommended routes and are not targeted. They do not have route recording functions, nor return route planning functions. The present invention refers to traditional route recommendation methods and proposes a personalized route recommendation method that combines user habits, driving trajectories, travel information and other information with personal preferences based on data analysis technology, deep learning technology, artificial intelligence technology, etc. The feature mining algorithm is used to mine the common route information of car owners to provide users with appropriate route information. According to the mined route information, the user history records are searched, and the user's travel habits are analyzed based on the user feature model. The route is recommended for the user, and the information such as the path recommendation results is input into the database. According to the user's needs, whether the route is recorded and saved for multiple use, and the return route is analyzed for correlation through the association method in the algorithm so that the user can call it directly.

[0003] Currently, traditional travel platforms mainly use some path planning algorithms to plan travel routes, which are not targeted. As people's living standards improve, users no longer just want a single route recommendation, but hope to get "exclusive" travel route recommendations that are more in line with their personal preferences. Summary of the invention

[0004] In order to overcome the problems of road conditions, environment, user suitability, etc. in reality, the present invention designs a personalized travel route recommendation method based on feature mining algorithm.

[0005] The personalized travel route recommendation method based on feature mining algorithm described in the present invention comprises the following steps:

[0006] Step 1: Collect basic user information and establish a user data information database;

[0007] Step 2: Analyze user information and establish a user feature model;

[0008] Step 3: Recommend personalized travel routes for users.

[0009] Furthermore,: the basic user information collected in step 1 includes recording the user's travel vehicle route, user personal information, and environmental information.

[0010] Furthermore, the specific process of the step 2, analyzing the user information and establishing the user feature model is:

[0011] Step 1: Divide the collected data information into m unrelated data blocks and send them to n nodes; generate the global candidate frequent item set by publicizing (1)

[0012]

[0013] Where: Op is the global candidate frequent item set; ∑z is the data formatting factor; m*n is the division of m into n unrelated data blocks; x is the local frequent item set;

[0014] The acquisition process of the above local frequent itemsets is: set the collected data as item set C = {A, B, C, D}, where A, B, C, D represent "time period", "weather conditions", "road conditions", "traffic information" and other information respectively, and each data set of A, B, C, D is divided into local data item sets, for example: A = {A1, A2, A3, ...A n}, where A1, A2, A3, and A4 represent "sunny", "cloudy", "foggy", "light rain", "heavy rain", and "snowy" respectively. Assuming the minimum support is 0.4, the local data set is scanned to obtain a local frequent data set such as: A x ={A1, A3, A5}, generate local frequent item set X = {A x , B x , C x , D x}.

[0015] Step 2: Calculate the support frequency of each candidate frequent item in the data through public (2), and arrange them in descending order of support;

[0016] Support(Ai=B)=P(Ai∪B) (2)

[0017] Where: the number of times each situation occurs is Ai, and the total database is B;

[0018] Step 3: Use the obtained header table and the sorted data to build a data set. According to the output results, generate relevant associations and segment each operation in the database. The calculation formula is:

[0019]

[0020] Where: P is the output result of the travel route information; Z represents the data that meets the conditions; S1 and S2 represent the travel route information, Represents the process of outputting travel route information.

[0021] Furthermore, the step 3 of recommending a personalized travel route for the user is to adjust the route and weight according to the travel destination, and change the recommendation algorithm result. The calculation formula is as follows:

[0022]

[0023] Where: Q represents user information; Kn represents user interest; g is the route recommendation weight, |z| is the optimization factor of the recommendation algorithm. Using the above formula, the calculation is repeated to generate new travel routes until a satisfactory result is generated.

[0024] Furthermore, there are two ways to record the user's travel route: the first is a manual recording method, in which the user directly records and saves the route that he or she has taken that meets his or her preferences through the system; the second is a machine recording method, in which the system detects the user's recent frequently taken routes and automatically records them.

[0025] Furthermore, the manually recorded route data is self-learned to analyze the user's driving habits and route preference characteristics, and the sparse autoencoder is used to extract the characteristics of the user's driving habits from a large amount of data to learn more implicit features of the data; the steps are as follows:

[0026] (1) Setting parameters such as learning rate, sparse parameter, number of iterations, number of neurons, randomly initializing weight w and bias b, extracting data features, and extracting four features: time, distance, traffic light, and road category from the manually recorded route data stored in the database;

[0027] (2) For the four feature quantities, forward pass them as the input of the sparse autoencoder to obtain the sparse expression of the intrinsic feature information of the four route feature quantities, and save the weight w1 and bias b1 of the sparse autoencoder after training;

[0028] (3) Initialize the input layer and hidden layer of the neural network with the obtained weight w1 and bias b1, reset the neural network, use the artificially recorded route data as training samples, train the neural network, extract features, and then train the next layer, layer by layer;

[0029] (4) Utilize the extracted training set features and send them to the feature classifier to achieve a feature classification; store the user features in the "interest database" so as to analyze the user's recorded routes through self-learning methods and extract features such as the user's personal habits and driving preferences.

[0030] Furthermore, the method for determining the return route includes the following steps:

[0031] (a) Determine the data to be analyzed, and determine the reference sequence of key feature information and the comparison sequence of feature information in the travel route; the data sequence reflecting the key feature information of the user's outbound route is called the reference sequence; the data sequence of factors affecting the return route recommendation is called the comparison sequence; let the reference sequence be Z = Z(k) | k = 1, 2...n; let the comparison sequence be X = X(k) | k = 1, 2...n;

[0032] (b) Due to the different physical meanings of various factors in the system, the dimensions of the route data may also be different, which makes it difficult to compare them. In order to reduce the difference in the absolute values ​​of the route data, they are unified into an approximate range and normalized. This paper adopts initialization processing, and the formula is as follows:

[0033]

[0034] Where K corresponds to the time period, and i corresponds to a row in the comparison series (i.e., a feature);

[0035] (c) Calculate the correlation coefficient. Let i be a fixed eigenvalue of the route data. That is to say, for a factor of a route data, each dimension is calculated to obtain a new data sequence. Each value in this sequence represents the correlation between the eigenvalue in the return route sequence and the corresponding dimension in the outbound route sequence (the larger the number, the stronger the correlation). The formula is as follows:

[0036]

[0037] (d) Calculation of correlation degree forms a route feature correlation sequence. The correlation coefficient is the correlation degree value between the key feature information sequence of the outbound route and the recommended factor sequence of the return route at each time. Therefore, it has more than one value. Therefore, the correlation coefficients of multiple times are concentrated, that is, the average value is calculated as the quantitative representation of the key feature information sequence of the outbound route and the recommended factor sequence of the return route. The formula is as follows:

[0038]

[0039] Sorting by correlation degree: Using the above formula, the calculated route feature information is sorted by correlation degree to find out the optimal correlation data information of each key feature point in the outbound route. The greater the correlation degree, the more similar the recommended return route is to the outbound route standard.

[0040] Beneficial effects of the present invention: The load balancing method based on SDN multi-controller improved ant colony provided by the present invention combines data analysis technology and deep learning technology, explores user interest characteristics as much as possible based on data mining algorithm, improves the personalization and satisfaction of recommendation results, adds the traveled routes to the search results, and then conducts deep learning through user feedback, so as to obtain path planning results that better meet user needs. At the same time, it can not only record and save the travel routes that users have traveled, so that users can use them directly when they travel next time, but also directly push out the return route according to the analysis of the personalized outbound route finally recommended by the system when preparing for the return trip, so that users can use it directly when they return.

[0041] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the travel route recommendation process.

[0043] Figure 2 This is a schematic diagram of different travel behaviors caused by the same traffic information.

[0044] Figure 3 This is a schematic diagram of sparse autoencoder training data.

[0045] Figure 4 It is a route planning process diagram.

[0046] Figure 5 It is a process diagram of correlation analysis return planning.

[0047] Figure 6 It is a flow chart of route planning method.

[0048] Figure 7 This is a diagram of the map function.

[0049] Figure 8 It is a route diagram.

[0050] Fig. 9 This is a recommended route diagram.

[0051] Fig.10 This is a recommended return route diagram.

[0052] Fig.11 This is a recommended return route diagram.

[0053] Fig.12 It is a manual route map. DETAILED DESCRIPTION

[0054] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose, the specific implementation methods, structural features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and embodiments.

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0056] In the description of the present invention, it is necessary to understand that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "aligned", "overlap", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0057] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features; in the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0058] Example 1

[0059] like Figures 1 to 4 As shown in FIG. 1 , the personalized travel route recommendation method based on the feature mining algorithm includes the following steps:

[0060] Step 1: Collect basic user information and establish a user data information database;

[0061] Step 2: Analyze user information and establish a user feature model;

[0062] Step 3: Recommend personalized travel routes for users.

[0063] Furthermore, the basic user information collected in step 1 includes recording the user's travel vehicle route, user personal information, and environmental information. The user personal information mainly refers to the user's age, gender, occupation, vehicle category, vehicle type, etc., and the environmental information mainly refers to the travel time period, weather conditions, road conditions, etc.; the basic information and necessary attributes on the travel platform are analyzed using big data analysis technology, and the data is divided according to the basic information of the platform and the basic information of the route that the user has recorded and saved, such as the travel time period is divided into "weekends" and "weekdays"; the weather conditions include "sunny", "cloudy", "foggy", "light rain", "heavy rain", "snowy", etc., and the road conditions are divided into "distance length", "time", "toll size", "number of traffic lights", etc. Based on the above data, a local frequent item set is generated.

[0064] Furthermore, the specific process of the step 2, analyzing the user information and establishing the user feature model is:

[0065] Step 1: Divide the collected data information into m unrelated data blocks and send them to n nodes; generate the global candidate frequent item set by publicizing (1)

[0066]

[0067] Where: Op is the global candidate frequent item set; ∑z is the data formatting factor; m*n is the division of m into n unrelated data blocks; x is the local frequent item set;

[0068] On this basis, the data is analyzed again to extract the value of low-density information on travel route information, including the number of events in various different situations in weather conditions, road conditions, traffic conditions, and historical road conditions. The travel route data is scanned, and the support frequency of each candidate data in the calculation data is established. The different situation information in these conditions is compared with the data in the total database.

[0069] Step 2: Calculate the support frequency of each candidate frequent item in the data through public (2), and arrange them in descending order of support;

[0070] Support(Ai=B)=P(Ai∪B) (2)

[0071] Where: the number of times each situation occurs is Ai, and the total database is B;

[0072] Then all frequent item sets that meet the conditions are output, the first frequent item set is put into the item header table, and they are arranged in descending order of support.

[0073] Step 3: Use the obtained header table and the sorted data to build a data set. According to the output results, generate relevant associations and segment each operation in the database. The calculation formula is:

[0074]

[0075] Where: P is the output result of the travel route information; Z represents the data that meets the conditions; S1 and S2 represent the travel route information, Represents the process of outputting travel route information.

[0076] Through the above process, the frequent item sets in the original data are arranged in descending order according to the support, and the mining of the characteristic information of the user's frequently traveled routes is completed.

[0077] Recommend the travel route information that the user is interested in to the user based on the user's characteristics. When the user starts searching for travel routes, optimize the search results and send more accurate content to the user. And calculate the recommended travel route based on the output content of the recommended route results plus the records in the interest database.

[0078] Furthermore, the step 3 of recommending a personalized travel route for the user is to adjust the route and weight according to the travel destination, and change the recommendation algorithm result. The calculation formula is as follows:

[0079]

[0080] Where: Q represents user information; Kn represents user interest; g is the route recommendation weight, |z| is the optimization factor of the recommendation algorithm. Using the above formula, the calculation is repeated to generate new travel routes until a satisfactory result is generated.

[0081] Furthermore, there are two ways to record the user's travel route: the first is a manual recording method, in which the user directly records and saves the route that he or she has taken that meets his or her preferences through the system; the second is a machine recording method, in which the system detects the user's recent frequently taken routes and automatically records them.

[0082] Furthermore, the manually recorded route data is self-learned to analyze the user's driving habits and route preference characteristics, and the sparse autoencoder is used to extract the characteristics of the user's driving habits from a large amount of data to learn more implicit features of the data; the steps are as follows:

[0083] (1) Setting parameters such as learning rate, sparse parameter, number of iterations, number of neurons, randomly initializing weight w and bias b, extracting data features, and extracting four features: time, distance, traffic light, and road category from the manually recorded route data stored in the database;

[0084] (2) For the four feature quantities, forward pass them as the input of the sparse autoencoder to obtain the sparse expression of the intrinsic feature information of the four route feature quantities, and save the weight w1 and bias b1 of the sparse autoencoder after training;

[0085] (3) Initialize the input layer and hidden layer of the neural network with the obtained weight w1 and bias b1, reset the neural network, use the artificially recorded route data as training samples, train the neural network, extract features, and then train the next layer, layer by layer;

[0086] (4) Utilize the extracted training set features and send them to the feature classifier to achieve a feature classification; store the user features in the "interest database" so as to analyze the user's recorded routes through self-learning methods and extract features such as the user's personal habits and driving preferences.

[0087] Furthermore, when the user obtains a satisfactory route recommendation, the system performs a round-trip correlation analysis based on the outbound route. Once the frequent item set of the outbound route is found, the algorithm can be used to generate associations based on the confidence of the outbound route. Find association rules such as "point A to point B" from the recommended route data set, that is, if the user goes from the departure point A to the destination point B, then the user is likely to need to return from point B to point A, and a new return route is needed. First, by extracting key information from the recommended outbound route, such as forks, highway entrances and exits, single and double lanes, etc., according to the correlation analysis, find the corresponding route information.

[0088] like Figure 5 As shown, the method for determining the return route includes the following steps:

[0089] (a) Determine the data to be analyzed, and determine the reference sequence of key feature information and the comparison sequence of feature information in the travel route; the data sequence reflecting the key feature information of the user's outbound route is called the reference sequence; the data sequence of factors affecting the return route recommendation is called the comparison sequence; let the reference sequence be Z = Z(k) | k = 1, 2...n; let the comparison sequence be X = X(k) | k = 1, 2...n;

[0090] (b) Due to the different physical meanings of various factors in the system, the dimensions of the route data may also be different, which makes it difficult to compare them. In order to reduce the difference in the absolute values ​​of the route data, they are unified into an approximate range and normalized. This paper adopts initialization processing, and the formula is as follows:

[0091]

[0092] Where K corresponds to the time period, and i corresponds to a row in the comparison series (i.e., a feature);

[0093] (c) Calculate the correlation coefficient. Let i be a fixed eigenvalue of the route data. That is to say, for a factor of a route data, each dimension is calculated to obtain a new data sequence. Each value in this sequence represents the correlation between the eigenvalue in the return route sequence and the corresponding dimension in the outbound route sequence (the larger the number, the stronger the correlation). The formula is as follows:

[0094]

[0095] (d) Calculation of correlation degree forms a route feature correlation sequence. The correlation coefficient is the correlation degree value between the key feature information sequence of the outbound route and the recommended factor sequence of the return route at each time. Therefore, it has more than one value. Therefore, the correlation coefficients of multiple times are concentrated, that is, the average value is calculated as the quantitative representation of the key feature information sequence of the outbound route and the recommended factor sequence of the return route. The formula is as follows:

[0096]

[0097] Sorting by correlation degree: Using the above formula, the calculated route feature information is sorted by correlation degree to find out the optimal correlation data information of each key feature point in the outbound route. The greater the correlation degree, the more similar the recommended return route is to the outbound route standard.

[0098] Through the above return route correlation recommendations, a return route that meets the user's personal preferences is recommended. When the user needs to return, the system directly pushes the planned route to the user for use, thereby obtaining a travel route that better meets the user's needs.

[0099] In addition, it should be noted that the user trajectory information is obtained by recording a large amount of trace point information, connecting the trace points, and finally obtaining the orbit information. In order to realize the path recording function, positioning is required first. Positioning refers to determining our position. Only when the positioning is accurate can the route taken be accurately recorded. Therefore, this article adopts the satellite positioning method. The advantage of satellite positioning is that it can be observed in real time and has low requirements for weather conditions. At least three satellites transmit electromagnetic waves to the target, obtain the distance between the satellite and the target, and then obtain the position of the satellite in the same coordinate system. In space, in addition to determining the coordinates of the three satellites, the distance from the satellite to the target can also be obtained, so the coordinates of the target can be obtained. The trace point information of the target is thus obtained. The satellite positioning module in the mobile phone is used to complete the satellite search and tracking and positioning speed and other data collection work. When the satellite positioning module collects the user's location information, positioning speed and other information, by constantly refreshing the electronic map, the user's position on the electronic map is displayed as a constantly changing trajectory.

[0100] The route point information is recorded according to the satellite positioning time, and finally the points are connected into a line to record and save the driving trajectory route. And it is stored in the historical database. When the route needs to be recorded, the historical trajectory points of this time period are obtained from the background, the first point is used as the starting point, and the last point is used as the end point. Combined with the target trace points obtained by the satellite system on the way, the driving trajectory of the route is drawn.

[0101] In summary, if Figure 6 As shown in FIG. 1 , the workflow of the load balancing method for improving ant colonies based on SDN multi-controllers is:

[0102] (1) First, determine the departure location A and destination location B on the platform.

[0103] (2) The system detects whether the route is one that has been traveled before, and prompts the user whether to select the route recorded in Solution 1 as the travel route. Solution 2 plans a route that is more in line with the user's personal preferences.

[0104] (3) If you refuse to select the previously recorded route, then use Solution 2 to generate a route plan from starting location A to destination location B.

[0105] (4) After selecting Option 2, the route planning is adjusted accordingly based on the online traffic condition monitoring data. The navigation provider may provide some suboptimal routes and add them to the search results for users to choose from. Then, the algorithm and user feedback are used for learning to obtain a route planning result that better meets user needs. A faster and more suitable road is determined in real time.

[0106] (5) Implement navigation based on the adjusted driving route planning.

[0107] (6) When the user arrives at the destination and needs to return, the system performs a correlation analysis on the outbound route, plans a return route that suits the user's personal preferences, and pushes it directly to the user.

[0108] This load balancing method based on SDN multi-controller improved ant colony combines data analysis technology and deep learning technology, and uses data mining algorithms to explore user interest characteristics as much as possible, improve the personalization and satisfaction of recommendation results, add the traveled routes to the search results, and then conduct deep learning through user feedback to obtain path planning results that better meet user needs. At the same time, it can not only record and save the travel routes that users have traveled, so that users can use them directly on their next trip, but also directly push out the return route based on the analysis of the personalized outbound route finally recommended by the system when preparing for the return trip, so that users can use it directly on the return trip.

[0109] Example 2

[0110] Since the travel route planning method designed to favor personal preferences is platform-independent, we can choose any travel platform such as Amap, Baidu Map, Tencent Map, etc. as the implementation of this detection system. In practical applications, by writing algorithms and porting programs to different platforms, the theory can be turned into practical applications, which will truly improve the suitability of travel route planning.

[0111] The open source OpenGIS and WebGIS specifications are applied to the tourism travel information query platform to realize the visualization operation of travel spatial geographic data; the open source JavaScript map engine maptalks.js in OpenGIS is used to improve users' overall experience of travel data on the map.

[0112] By referencing the digital map JavaScript API, the data is displayed on the street map of the digital map through point marks through JavaScript script code, and the comprehensive information of the area to be analyzed is obtained. Based on the comprehensive information, the key feature point information of the area to be analyzed is extracted. Secondly, the user's travel data is obtained from the historical database, and the user's corresponding travel route is calculated based on the travel data, and the recommended route matching is performed based on the user feature model. The generated results are analyzed to see whether the route is of interest to the user, thereby achieving effective analysis of a large amount of travel data, and then providing reliable analysis results for traffic planning, thereby improving the utilization rate of traffic resources.

[0113] When the user is ready to travel, he / she enters the starting point and destination in the travel platform. After the user searches for the route, he / she calls the electronic map. The street view display and 3D map will feed back the information to the user. The displayed route recommends several driving routes. After the user registers and fills in personal information in the travel platform, the travel routes selected by users with different characteristics are different. The personal information is collected and entered into the database. Combined with the routes and driving patterns recorded by the user in recent period of time, when the user needs to travel, a route that meets the user's characteristics and preferences is recommended, such as from "Weiyang Campus of Xi'an University of Technology" to "Xi'an Qinling Wildlife Park". Figure 7 shown.

[0114] Xi'an Qinling Wildlife Park is a tourist attraction located in a remote area, with a total distance of nearly 70 kilometers, which is far from the departure point. It is necessary to consider "weather conditions", "distance length", "traffic light conditions", "toll gates", "road congestion" and other conditions. On weekdays, the traffic pressure to this place is relatively small, but when the number of people choosing to travel increases during holidays, the traffic volume to the zoo will increase dramatically, and this section of the road is in a state of traffic congestion most of the time. The recent journey of user A to the tourist attraction and the manually recorded route conditions are entered into the database as reference samples for feature extraction, and then a travel route to the tourist attraction that meets the user's characteristics is recommended.

[0115] Assume that user A chooses to travel to tourist attractions on weekdays. The roads to tourist attractions are relatively smooth, the number of tourists in tourist attractions is small, and there is plenty of time to play in the scenic spots, reducing the waiting time in line. During the journey, the user is more inclined to choose roads with less tolls and more traffic lights, rather than routes with less time and more costs. Figure 8 As shown by the red route, there is no toll station on this road section. It is prone to congestion during rush hour and takes a long time. It is most suitable for users to travel during non-rush hour on weekdays, which is in line with the user's driving characteristics.

[0116]

[0117] When users travel during holidays, the roads to tourist attractions are congested, the number of tourists in scenic spots is large, and the time for sightseeing is reduced. Therefore, user A tends to choose roads that avoid congestion and reduce waiting time, choose roads with high tolls and few traffic lights, try to avoid main roads, and choose some small roads on the route to detour in order to save time. This route avoids the main road and detours to the toll highway, which greatly reduces the travel time and is less likely to cause congestion. Fig. 9 The red route is shown. The recent journey of user A to tourist attractions is entered into the database as a reference sample for classification. The elements such as "time period", "road conditions", and "traffic conditions" in the database are extracted through the feature mining algorithm. The input is mapped to a value of (0,1) using the feature classifier. The following formula is used

[0118]

[0119] The probability values ​​of route features selected by users in different situations and time periods are calculated. When recommending routes, the system selects the route with the highest probability of user selection at that time and recommends it to the user as a personalized recommended route based on the user's characteristics. When user A starts searching for travel routes, the route and weight are adjusted based on the travel route recommendation results and the probability values ​​of route features in the database to generate route recommendations. The calculation formula is as follows:

[0120] Using the above formula, the calculation is repeated to generate new travel routes until a satisfactory result is generated.

[0121] When user A returns, we conduct a correlation analysis based on the route we came from. First, we determine the key feature information in the travel route. When the user returns, the time of the return trip has changed, and the situation of the driving route has also changed accordingly. The route features of the outbound trip have been stored. Let the data sequence of the key feature information of the user's outbound route be the reference sequence. When returning, we calculate the correlation coefficient, that is, we calculate the characteristic situation of this trip data to obtain a new data sequence. Each value in this sequence represents the correlation between the characteristic value in the return route sequence and the corresponding dimension in the outbound route sequence. We perform correlation calculation to form a route feature correlation sequence. The correlation coefficient is the correlation degree value between the key feature information sequence of the user's outbound route and the recommended factor sequence of the return route at each moment, using the following formula:

[0122]

[0123] Assume that user A goes to Xi'an Qinling Wildlife Park on a weekday and returns in the evening. This is the rush hour and the traffic volume on the main roads and highways in the city increases. At this time, try to choose the alleys in the city for the return trip to avoid congested routes. The return route is recommended based on the return route and the user's characteristics and preferences. This route uses the small roads and alleys in the city to alleviate congestion time. Fig.10 As shown in the red route, if the user goes to Xi'an Qinling Wildlife Park during holidays and has plenty of time, they are more likely to choose a route with less tolls to return, try to avoid the alleys in the city, and choose the main roads. Fig.11 Route shown in red.

[0124] When user A is traveling and feels satisfied with the route he has driven or some "small roads" he has traveled based on his personal driving experience, he can choose to save it directly, extract the trace point information of the route recorded at the time of satellite positioning, and finally make a line from the points to record and save the driving trajectory route. When the user travels next time, he can directly call the travel route recorded and saved in the database for direct use, such as Fig.12 The black route is shown.

[0125] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A personalized travel route recommendation method based on feature mining algorithm, characterized in that: The steps include: Step 1: Collect basic user information and establish a user data information database; Step 2: Analyze user information and establish a user feature model; the specific process is: Step 1: Divide the collected data information into m unrelated data blocks and send them to n nodes; generate the global candidate frequent item set by publicizing (1) Where: Let Op be the global candidate frequent item set; ∑z is the data formatting factor; m*n means dividing m into n irrelevant data blocks; x is the local frequent item set; Step 2: Calculate the support frequency of each candidate frequent item in the data through public (2), and arrange them in descending order of support; Support(Ai=B)=P(Ai∪B) (2) Where: the number of times each situation occurs is Ai, and the total database is B; Step 3: Use the obtained header table and the sorted data to build a data set. According to the output results, generate relevant associations and segment each operation in the database. The calculation formula is: Where: P is the output result of the travel route information; Z represents the data that meets the conditions; S1 and S2 represent the travel route information, Represents the process of outputting travel route information; Step 3: Recommend personalized travel routes for users. Specifically, the user's travel route data is used to adjust the route and weight, and change the recommendation algorithm result. The calculation formula is as follows: Where: Q represents user information; Kn represents user interest; g is the route recommendation weight, |z| is the optimization factor of the recommendation algorithm. Using the above formula, the calculation is repeated to generate new travel routes until a satisfactory result is generated. The method for determining the return route includes the following steps: (a) Determine the data to be analyzed, and determine the reference sequence of key feature information and the comparison sequence of feature information in the travel route; the data sequence reflecting the key feature information of the user's outbound route is called the reference sequence; the data sequence of factors affecting the return route recommendation is called the comparison sequence; let the reference sequence be Z = Z(k) | k = 1, 2...n; let the comparison sequence be X = X(k) | k = 1, 2...n; (b) Due to the different physical meanings of various factors in the system, the dimensions of the route data may also be different, which makes it difficult to compare them. In order to reduce the difference in the absolute values ​​of the route data, they are unified into an approximate range and normalized. This paper adopts initialization processing, and the formula is as follows: Where K corresponds to the time period, and i corresponds to a row in the comparison series, i.e. a feature; (c) Calculate the correlation coefficient. Let i be a fixed eigenvalue of the route data. That is to say, for a factor of a route data, each dimension is calculated to obtain a new data sequence. Each value in this sequence represents the correlation between the eigenvalue in the return route sequence and the corresponding dimension in the outbound route sequence. The larger the number, the stronger the correlation. The formula is as follows: (d) Calculation of correlation degree forms a route feature correlation sequence. The correlation coefficient is the correlation degree value between the key feature information sequence of the outbound route and the recommended factor sequence of the return route at each time. Therefore, it has more than one value. Therefore, the correlation coefficients of multiple times are concentrated, that is, the average value is calculated as the quantitative representation of the key feature information sequence of the outbound route and the recommended factor sequence of the return route. The formula is as follows: Sorting by correlation degree: Using the above formula, the calculated route feature information is sorted by correlation degree to find out the optimal correlation data information of each key feature point in the outbound route. The greater the correlation degree, the more similar the recommended return route is to the outbound route standard.

2. The personalized travel route recommendation method based on feature mining algorithm according to claim 1, characterized in that: The basic user information collected in step 1 includes recording the user's travel vehicle route, user personal information, and environmental information.

3. The personalized travel route recommendation method based on feature mining algorithm as claimed in claim 2, characterized in that: There are two ways to record the user's travel route: the first is a manual recording method, in which the user directly records and saves the route that he or she has taken that meets his or her preferences through the system; the second is a machine recording method, in which the system detects the user's recent frequently taken routes and automatically records them.

4. The personalized travel route recommendation method based on feature mining algorithm as claimed in claim 3, characterized in that: The manually recorded route data is self-learned to analyze the user's driving habits and route preference characteristics, and the sparse autoencoder is used to extract the characteristics of the user's driving habits from a large amount of data to learn more implicit features of the data; The steps are as follows: (1) Setting parameters such as learning rate, sparse parameter, number of iterations, number of neurons, randomly initializing weight w and bias b, extracting data features, and extracting four features: time, distance, traffic light, and road category from the manually recorded route data stored in the database; (2) For the four feature quantities, forward pass them as the input of the sparse autoencoder to obtain the sparse expression of the intrinsic feature information of the four route feature quantities, and save the weight w1 and bias b1 of the sparse autoencoder after training; (3) Initialize the input layer and hidden layer of the neural network with the obtained weight w1 and bias b1, reset the neural network, use the artificially recorded route data as training samples, train the neural network, extract features, and then train the next layer, layer by layer; (4) Using the extracted training set features, the features are sent to the feature classifier to achieve a feature classification; the user features are stored in the "interest database", so as to analyze the user's recorded routes through self-learning methods and extract features such as the user's personal habits and driving preferences.