CN113239269A

CN113239269A - Improved recommendation algorithm-based travel route customization method

Info

Publication number: CN113239269A
Application number: CN202110483180.3A
Authority: CN
Inventors: 钱霖奕
Original assignee: Hohai University HHU
Current assignee: Hohai University HHU
Priority date: 2021-04-30
Filing date: 2021-04-30
Publication date: 2021-08-10

Abstract

The invention discloses a tour route customizing method based on an improved recommendation algorithm, which comprises a tour APP installed in a mobile phone or a computer of a user, wherein the tour APP has two modes, namely a common mode and a free mode; in a common mode, a user selects and inputs or does not input a starting point, and after an end point is input, the travel APP provides six routes for the user, wherein the six routes are respectively the shortest route, the most scenic spots, the most food, the optimal price, the least pedestrian volume and the lowest poor rating route; in the follow-up mode, the travel APP automatically obtains the starting point of the user and generates a random end point according to an improved recommendation algorithm. The invention can provide customized route service for the user by improving the recommendation algorithm and data analysis, is closer to the user requirement, and has simple tour APP interface and convenient use. The tour APP makes an innovative scoring system aiming at the phenomenon that software is commented on the market at present, namely poor comment of a user is supported, good comment service is not provided, and the phenomenon that part of user experience is poor due to good comment is effectively improved.

Description

Improved recommendation algorithm-based travel route customization method

Technical Field

The invention belongs to the technical field of dynamic route planning, and particularly relates to a travel route customizing method based on an improved recommendation algorithm.

Background

The existing map software can only provide traditional routes such as taxi taking, driving, bus transportation, walking and the like. With diversification of user demands, the APP cannot provide a more humanized travel route, such as a route with the most food, the least people flow, the lowest bad comment and the like.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, provides map software which can provide route customization service for users based on user preference and can carry out dynamic route planning, and becomes an instant short-distance travel assistant for urban crowds.

The invention is realized by the following technical scheme:

a travel route customizing method based on an improved recommendation algorithm comprises a travel APP installed in a mobile phone or a computer of a user, the user enters the travel APP through registration and login, and the user fills in age, occupation, gender, academic calendar, hobbies, travel frequency, travel budget and route preference during registration. The travel APP has two modes for a user to select, namely a common mode and a free mode;

in a common mode, a user selects and inputs or does not input a starting point, and after an end point is input, the travel APP provides the most scenic spots, the shortest route, the most food, the optimal price, the least pedestrian volume and the least poor rating route for the user;

in the follow-up mode, the travel APP automatically obtains the starting point of the user and generates a random end point according to an improved recommendation algorithm.

In the invention, the travel APP uses a Baidu map API interface to obtain the distance between a starting point and a destination point, the geographic information, the business hours, the pictures, the prices and the like of scenic spots, shops and restaurants in an effective area, and the Ali cloud database stores the information of each user, an algorithm model and information contained in a bad comment system.

After the route is generated by the route planning model, a user can view basic information such as addresses, pictures, prices, evaluations and business hours by clicking coordinates on the route, and aiming at the existing evaluation systems such as popular comment, American group, hundred-degree maps and the like, a plurality of users can reflect that a plurality of shops have different services and scores, namely, part of restaurants can be scored manually to result in the condition of high score, and aiming at the point, the evaluation system of the travel APP adopts a brand-new 'poor score' mode to score, so that the user can make a better choice. Specifically, the APP only supports poor comment of the user and does not provide good comment service, the phenomenon that the bad comment is refreshed to cause unpleasant experience of part of users can be effectively improved, meanwhile, the APP specifies that the poor comment of the user needs to upload evidences such as pictures and videos, and therefore the phenomenon that the bad comment is refreshed maliciously is suppressed. On the basis of a poor evaluation system, the APP provides a path with the lowest poor evaluation ratio, the poor evaluation quantity and the store flow are used as a ratio to obtain the poor evaluation ratio, and the path with the lowest poor evaluation ratio is generated according to the poor evaluation quantity and the store flow.

In the invention, the algorithm model comprises a route planning model and an improved recommendation algorithm model, the improved recommendation algorithm model adopts the weighting calculation of the label similarity of the user and the similarity calculated by the collaborative filtering algorithm based on the user, 50% of weight is assigned to each improved recommendation algorithm model, and the number of parameters related to the label similarity of the user is 8, including age, occupation, gender, academic calendar, hobbies, travel frequency, travel budget and route preference.

Improved recommendation algorithm model: the conventional user-based collaborative filtering algorithm is to calculate the similarity between users by using the feedback of the users to the articles, and after improving the recommendation algorithm model, weighting calculation (each is weighted by 50%) is carried out on the label similarity of the users and the similarity calculated by the user-based collaborative filtering algorithm so as to carry out more accurate matching and comparison.

The number of parameters related to the similarity of the user tags is 8, and a parameter f is assumed_iHas a weight of w_iThen the user label Similarity₁The calculation method of (2) is as follows:

Similarity₁＝∑f_i*w_i

the weight distribution is obtained in the early stage of APP production by adopting a questionnaire form, and the results are shown in the following table 1:

TABLE 1

Parameter(s)	Weight value
		Age	0.15
Occupational occupancy	0.08
		Sex Gender genter	0.11
Calendar Edutation	0.12
		Hobbies	0.28
Travel frequency Travelfequency	0.08
		Travel budget TravelBudge	0.11
Route preference RoutePreference	0.07

In summary, if the Similarity calculated based on the collaborative filtering algorithm of the user is Similarity₂Then, the Similarity calculated by the improved recommendation algorithm is:

Similarity＝0.5*Similarity₁+0.5*Similarity₂＝0.5*∑f_i*w_i+0.5*Similarity₂

a route planning model: and constructing an undirected graph G < V, E > according to the starting point S and the end point T of the user and various coordinate points in the range, and planning a proper route for the undirected graph according to the route preference selected by the user. Assuming that the route preference selected by the user is K, which is a feature value of the route to be planned, such as food, pedestrian volume, poor rating, and the like, the feature value K needs to be transferred to a point weight of each coordinate point through model calculation, and then if the shortest distance from the starting point S to the terminal point T is satisfied, a map API interface can be called to obtain a side weight (distance) between the point and the point, so that the route planning algorithm can be abstracted as a graph theory algorithm. Planning a route from a starting point S where a user is located to an end point T is substantially the improved Dijkstra shortest path algorithm, namely, finding a path from the point S to the point T in the directed graph meets the requirements of edge weight sum minimum and point weight sum maximum.

The travel APP firstly calls a Baidu API to obtain coordinates from a starting point to an end point effective area, a mark array is set, if the coordinates are scenic spot marks 1, otherwise, the coordinates are 0, then the coordinates are weighted and summed with the matching degree, and the weight ratio is set to be a golden ratio

And after the point weight array is updated, calling a route planning model to obtain the most routes of the scenic spots.

The travel APP firstly calls a Baidu API to obtain coordinates in an effective area from a starting point to an end point, a mark array is set, if the coordinates are restaurants or snacks, the mark array is marked as 1, otherwise, the mark array is 0, then the coordinates are weighted and summed with the matching degree, and the weight ratio is set as a golden ratio

And calling a route planning model after updating the point weight array to obtain the route with the most food.

The tour APP firstly calls a Baidu API to obtain coordinates from a starting point to an end point effective area, the matching degree of the coordinates is calculated by using an improved recommendation algorithm model, then the coordinates with higher matching degree are preferentially screened, the coordinate points with lower matching degree are deleted, the point weight value array is updated to be a negative value of the per-capita consumption price, and a route planning model is used for generating an optimal price route after updating.

The tour APP firstly calls a Baidu API to obtain coordinates from a starting point to an end point effective area, the point weight of each coordinate point is updated to be the real-time pedestrian flow, a negative value is required when the point weight array is updated, and a route planning model is used for generating a pedestrian flow minimum route after the updating is finished.

In the invention, for the route provided by the travel APP, a user can click and check the address, the difference rating, the picture, the price and the comment information of the coordinate in the route.

Compared with the prior art, the invention has the beneficial effects that:

the invention can provide customized route service for the user through the improved recommendation algorithm and data analysis, is closer to the user requirement, and has simple tour APP interface and convenient use. The tourism APP makes an innovative scoring system aiming at the phenomenon that software is commented on the market at present, namely poor comment of a user is supported, good comment service is not provided, and the phenomenon that part of user experience is not good due to good comment is effectively improved.

Drawings

FIG. 1 shows the software architecture of travel APP;

FIG. 2 shows two basic mode work flow diagrams of the travel APP;

FIG. 3 shows an example of a free-wheeling mode;

FIG. 4 shows an algorithmic flow chart for the free mode;

FIG. 5 shows an example of a user-based collaborative filtering algorithm;

FIG. 6 shows a graph of shortest path effects;

FIG. 7 shows an algorithm flow for the most routes of the sights;

FIG. 8 shows a view of the most course effect of the sights;

FIG. 9 shows an algorithm flow for the food majorities route;

FIG. 10 shows the effect of the maximum route of food;

FIG. 11 shows an algorithm flow for a price optimized route;

FIG. 12 shows an algorithm flow for a least pedestrian traffic route;

FIG. 13 shows a basic framework of a review system;

fig. 14 shows an example of a poor scoring system.

Detailed Description

As shown in FIG. 1, the overall architecture of software is shown, and the basic architecture of the invention is two models of data processing, improved recommendation algorithm and route planning, and an evaluation system is established on the basis of the two models.

The data processing comprises data acquisition and storage.

Besides user information and an algorithm model, data acquisition also needs to dynamically call a Baidu map API interface to obtain the following data:

effective area: firstly, the distance S between a starting point and an end point is obtained, then the starting point is used as the circle center, the radius is S to make a circle, the end point is used as the circle center, the radius is S to make a circle, and the intersected part of the two circles is the effective area (the coordinate points selected by us are all contained in the area).

The following parameters are obtained for scenic spots, shops, restaurants and the like in the effective area:

geographic Position

Distance from the starting point

Business hours BusinessHours

Pictures

Price

The data is temporarily stored in a database to provide data for route planning in the latter two modes.

The data storage comprises short-term storage and long-term storage, the user data and the data of each shop, scenic spot and restaurant in the poor evaluation system need to be stored for a long term, and the data can be deleted only when the user logs off the account or does not log in the account for a long term or the shops, scenic spots and the like at a certain position are permanently closed; the short-term storage aims at dynamically acquired data, such as parameter information in a range acquired by calling an API (application programming interface) by a user for one route planning, and the parameter information is deleted after the user arrives at a destination or closes the APP so as to reduce space expenditure.

As shown in FIG. 2, travel APP has two main modes: a normal mode and a free mode.

In a common mode, the APP defaults that the user has a destination of the mental apparatus, the user can select to input or not input a starting point, after the end point is input, the software can generate a plurality of routes according to the selection of the user, the routes comprise a plurality of routes with the most food, the most scenic spots, the shortest distance and the like, and after the user selects one route, the APP can customize the route for the APP by utilizing the two models.

In consideration of the fact that in real life, many people who arrive at a city often do not know where to play because of the unfamiliarity of life, APP increases the casual mode. In the "follow-up" mode, the APP will generate a random destination according to the recommendation algorithm, and when the user reaches the destination, the APP will continue to generate the next destination. In this mode, the user can give the software the option of being careful to experience a pleasant feeling of opening the blind box.

In the random mode, since the APP specifies the destination, an appropriate radius must be defined to reduce the complexity, for example, the set threshold is 1.5km to 2km, that is, each time the destination is generated, an appropriate destination is selected from a circle with a radius of 1.5km to 2km, which is centered around the location of the user, the selection rule is the matching degree between the user and the coordinate point, as shown in fig. 3, and the matching degree of seven points near the starting point satisfies the condition shown in table 2 below:

TABLE 2

Numbering	Degree of matching
		A	0.5
B	0.34
		C	0.33
D	0.54
		E	0.36
F	0.47
		G	0.45

The coordinate D can be determined to be more in line with the preference of the user by using a simple sorting algorithm, so D is set as the next destination. And repeating the steps after the user reaches the destination until the user exits the free mode, wherein the algorithm flow of the free mode is as shown in figure 4.

The improved recommendation algorithm model comprises a conventional collaborative filtering algorithm based on a user and a user label similarity algorithm.

The algorithm of the conventional user-based collaborative filtering algorithm is described as follows:

given a user u and a user v, let N (u) denote the item set where u has positive feedback, and let N (v) denote the item set where v has positive feedback, we can calculate the interest similarity of u and v by the Jaccard formula:

or by cosine similarity:

as shown in fig. 5, if the user a has positive feedback on the item { a, C }, the user B has positive feedback on the item { B, C }, and the user C has positive feedback on the item { a, C, D }, the cosine similarity calculation includes:

the similarity between the user A and the user B is as follows:

the similarity between the user A and the user C is as follows:

so that the similarity between the user a and the user C is higher, and the item D is recommended to the user a.

The user label similarity algorithm selects 8 parameters in total, which are detailed in the parameters in table 1, firstly, non-numerical parameters, such as gender, occupation, academic calendar and the like, are quantified, and because the magnitude of each parameter is different after quantification, the magnitude needs to be unified, and the process is as follows:

TABLE 3

For a certain user x, the ith parameter can be expressed as f_x，iThen a vector label V can be established_x＝(f_x，1，f_x，2，f_x，3，...，f_x，8) Finally, for two different users A and B, the vector label V is known_A＝(f_A，1，f_A，2，f_A，3，...，f_A，8) And V_B＝(f_B，1，f_B，2，f_B，3，...，f_B，8) Can calculate its Euclidean distance S_A，BNote that the arrangement shown in Table 1 above is addedThe weight of (c):

the route planning model provided by the invention uses an improved Dijkstra algorithm, and coordinate information in an effective area range is obtained in the data acquisition process, so that an undirected graph G & lt V, E & gt can be established, the side length of any two nodes in the graph is a distance, and the point weight of each node i in a vertex set V is assumed to be w_iThen, the following algorithm is used:

setting variables:

maximum number of nodes maxn

The point weight w [ maxn ]

The edge weight adjacency matrix G [ maxn ]

Starting point S

End point T

The shortest distance D [ maxn ] from the starting point S to other points

The sum of the maximum point weights W [ maxn ] from the starting point S to the other points

Node Access markers vis [ maxn ]

The predecessor node array of each node pre [ maxn ]

Infinite upper bound inf

The algorithm steps are described in detail as follows:

1. initialization: setting the distance from the starting point S to other points as inf, D [ S ] is 0, W [ S ] is W [ S ], pre [ S ] is-1

2. Traversing to find the point u with the shortest distance in the nodes which are not traversed currently

3. Marking node u, updating minimum value

4. And taking the node v as an intermediate node, and executing updating operation on all nodes which are not accessed:

5. when D [ u ] + G [ u ] [ v ] < D [ v ], D [ v ] ═ D [ u ] + G [ u ] [ v ], pre [ v ] ═ u, and W [ v ] ═ W [ u ] + W [ v ]

6. When D [ u ] + G [ u ] [ v ] ═ D [ v ], if W [ u ] + W [ v ] > W [ v ], W [ v ] ═ W [ u ] + W [ v ]

7. The DFS output path may be completed after the two cycles are completed.

And (3) algorithm complexity analysis:

the time complexity of the non-optimized shortest-path algorithm is O (N)²) N is the maximum node number, and obviously, the node number is slower under the condition of larger node number, so that heap optimization can be adopted to reduce the time complexity to O (NlogN).

All the following lines are planned based on the algorithm, and for the lines with different characteristics, only how to transfer the characteristic values into point weights needs to be considered, which is specifically described as follows:

the shortest distance

The most basic route in route planning, the point weight is the matching degree of the user and the coordinate point. Because the improved recommendation algorithm model in APP actually only calculates the degree of match between users. In order to obtain the matching degree required by the former, deductive reasoning is applied, namely the similarity between the users obtained by the improved recommendation calculation is directly regarded as the matching degree between the users and the articles, for example, the similarity between the user A and the user B is X, the user B has positive feedback on the article C, and the matching degree between the user A and the article C is also X. After the point weight is calculated, the shortest distance route can be obtained by directly applying a route planning model.

Fig. 6 is a diagram showing the effect of the shortest route.

② most scenic spots

When planning the most routes of the scenic spots, a marker array sensor [ maxn ] is started without planning a route which is all scenic spots]If the coordinate is the mark 1 of the scenic spot, otherwise, the coordinate is 0, then the coordinate is weighted and summed with the matching degree, because the characteristic of the route is the scenic spot, the weight ratio is set as the golden ratio

Namely, the original point weight array (based on the shortest distance) is updated as:

and calling a route planning model after the point weight array is updated to obtain the most routes of the scenic spots.

FIG. 7 is a flowchart of the most scenic spot route algorithm, and FIG. 8 shows an effect diagram of the most scenic spot route.

(iii) maximum food

Setting a mark array food [ maxn ], if the coordinate is restaurant or snack, marking as 1, otherwise, 0, and updating the point weight array when the ratio is regarded as golden ratio,

Fig. 9 is a flowchart of an algorithm of the food maximum route, and fig. 10 shows an effect diagram of the food maximum route.

Fourthly, route with optimal price

The route is intended to provide a most cost-saving travel scheme for the user, because the price is in a large order of magnitude, and the error is obviously large if the weighted sum is adopted to update the point weight group (on the premise of magnitude difference, the matching degree hardly contributes to the result), that is, the coordinates on the generated route may not accord with the preference of the user, so a route planning strategy is required at this time:

in the data acquisition process, coordinate information in an effective area is acquired, so that the matching degree can be calculated by using an improved recommendation algorithm model, then the coordinates with higher matching degree can be preferentially screened, and the coordinate points with lower matching degree can be directly deleted, so that the method has two benefits: firstly, the finally planned route must be more in line with the preference of the user, secondly, the complexity can be reduced after some nodes with lower matching degree are deleted, the time of model operation is shortened, and the route planning efficiency is improved.

After the nodes with lower matching degree are deleted, the point weight array is directly updated to the price per capita, the price optimal route is required, and therefore the point weight is updated to be negative:

W[i]＝-price[i]

after the updating, the route planning model can be used to generate the route with the optimal price, and fig. 11 shows an algorithm flow chart of the route with the optimal price.

The minimum flow of people

The route aims to provide a route with the least sum of real-time pedestrian volume of scenic spots, shops and restaurants on the path for the user. The point weight of each coordinate point is updated to be the real-time pedestrian traffic Foottraffic, but the maximum point weight path is obtained in the route planning model, so that the point weight group is updated by taking the following negative value:

W[i]＝-FootTraffic[i]

and planning the route by using the route planning model on the basis. The difficulty with this route planning is to obtain real-time traffic. For obtaining the real-time people flow of the scenic spot or the restaurant, the direct realization of software is not practical, so the solution of the invention is to cooperate with the scenic spot or the shop on the basis of a certain user, APP provides the user flow, and the APP provides the real-time people flow information. For restaurants with relatively small pedestrian volume, the current pedestrian volume information can be directly provided through the host computer at the front stage of the restaurants, and comprises the number of reserved persons and the number of current diners; for scenic spots with huge pedestrian flow, the pedestrian flow acquired by a single path is obviously inaccurate, so the following three ways are adopted:

I. counting channel gate

Gate equipment is installed at the entrance and the exit of the scenic spot, and the number of tourists entering and leaving the garden is accurately counted in real time in a one-man gate verification mode.

Ticket selling system

Each scenic spot has its own off-line or on-line ticketing system, but these data can only be used as a reference for the number of visitors entering the garden.

III. Infrared device

An infrared device is installed at the entrance and exit position of a scenic spot, the passenger flow volume is added by 1 when an object passes through the scenic spot, but the influence of the environment is caused, the moving object is difficult to distinguish whether the moving object is a person or an object, and the data accuracy is low.

For the three modes, the number of tourists in the garden is calculated by adopting a weighting method, and the accuracy rate of the number of tourists entering the garden is obvious: i is more than II and more than III, therefore, the number of people measured by the three methods is num by adopting the weighted calculation of 50 percent, 30 percent and 20 percent₁，num₂，num₃And then the number of people entering the garden is as follows:

N_into＝0.5*num₁+0.3*num₂+0.2*num₃

For the number of people leaving the garden, only the first method and the third method can be counted, so that if the weighted calculation is carried out for 80% and 20%, the following steps are carried out:

N_into＝0.8*num₁+0.2*num₃

The real-time number of people in the scenic spot is as follows:

N＝N_into-N_{Go out}

Note: the above weight assignment can be adjusted according to actual conditions, and is not an optimal solution.

The APP determines the pedestrian flow of each scenic spot and each restaurant, sets a real-time pedestrian flow number Foottraffic, updates the point weight number array, and finally generates a route by using a route planning algorithm.

Fig. 12 is a flow chart of the algorithm for the human traffic minimization route.

The above are all basic architectures of the APP, a comment system-poor comment system of the APP is established on the basis, and after a route is generated, a user clicks coordinates on the route to check specific information of the coordinates, and as shown in fig. 13, the comment system is a basic framework of the comment system.

As shown in FIG. 14, an example of a bad scoring system is shown, for a seafood cafeteria in Nanjing urban area, the total people flow counted by APP is 19873 people, the total bad scoring is 463, and the calculated bad scoring is calculated as bad scoring

And a bar graph is shown on the right side for the user to feel more intuitively.

The poor evaluation system enables route planning to be one more choice, namely the lowest poor evaluation route and the lowest same-person flow route, and the point weight value still needs to be updated negatively:

W[i]＝-FeedBackRate[i]

and then calling a route planning model to generate a route.

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered as the technical scope of the present invention, the technical solution of the present invention and the inventive concept thereof should be equally replaced or changed within the scope of the present invention.

Claims

1. A tourism route customization method based on improved recommendation algorithm is characterized by comprising a tourism APP installed in a mobile phone or a computer of a user, wherein the user enters the tourism APP through registration and login, and the tourism APP has two modes for the user to select, namely a common mode and a free mode;

in a common mode, a user selects and inputs or does not input a starting point, and after an end point is input, the travel APP provides a route with the shortest distance, the most scenic spots, the most delicious food, the optimal price, the least pedestrian volume and the lowest poor rating for the user;

2. The improved recommendation algorithm based travel route customization method according to claim 1, wherein the travel APP uses a Baidu map API interface to obtain the distance between a starting point and a destination point, geographical information of scenic spots, shops and restaurants in an effective area, the distance from the starting point, business hours, pictures and prices, and the Aliskiu database stores information of each user, an algorithm model and information contained in a poor scoring system.

3. The improved recommendation algorithm based travel route customization method according to claim 2, wherein the user fills in age, occupation, gender, academic calendar, hobbies, travel frequency, travel budget and route preference at registration.

4. The method as claimed in claim 3, wherein the algorithm model includes a route planning model and an improved recommendation algorithm model, the improved recommendation algorithm model is calculated by weighting both the similarity of the user's label and the similarity calculated by the collaborative filtering algorithm based on the user, each of the improved recommendation algorithm model is weighted by 50%, and the similarity of the user's label is related to 8 parameters including age, occupation, gender, academic calendar, hobby, frequency of travel, budget of travel, and route preference.

5. The method for customizing a travel route based on an improved recommendation algorithm as claimed in claim 4, wherein the route planning model is an improved Dijkstra shortest-path algorithm, specifically: constructing an undirected graph according to a starting point and an end point of a user and coordinate points in a range, then according to the preference of a route selected by the user, wherein the preference of the route is a characteristic value of the line to be planned, transferring the characteristic value into a point weight of each coordinate point through model calculation, then calling a map API (application program interface) to obtain an edge weight between the point and the point if the distance from the starting point to the end point is shortest, and obtaining the edge weight between the point and the point if the sum of the edge weights is minimum, wherein the route with the largest sum of the point weights is the route planned by the user.

6. The method as claimed in claim 5, wherein the travel APP calls the Baidu API to obtain the coordinates from the start point to the end point within the effective area, sets a mark array, if the coordinates are the sight mark 1, otherwise, the coordinates are 0, then weights and sums with the matching degree, and sets the weight ratio as the golden ratio

7. The improved recommendation algorithm based travel route customization method according to claim 5, wherein the travel APP first calls the Baidu API to obtain the coordinates from the start point to the end point within the effective area, sets a mark array, marks the coordinates as restaurant or snack if the coordinates are 1, otherwise, marks the coordinates as 0, then performs weighted summation with the matching degree, and sets the weight ratio as golden ratio

8. The method as claimed in claim 5, wherein the travel APP first calls a Baidu API to obtain coordinates from a start point to an end point within an effective area, the improved recommendation algorithm model is used to calculate the matching degree of the coordinates, then the coordinates with higher matching degree are preferentially selected, the coordinate points with lower matching degree are deleted, the point weight array is updated to a negative value of the per-capita consumption price, and after the updating, the route planning model is used to generate the route with the optimal price.

9. The method as claimed in claim 5, wherein the travel APP first calls a Baidu API to obtain coordinates from a starting point to an end point within an effective area, the point weight of each coordinate point is updated to a real-time pedestrian volume, a negative value is taken when the point weight group is updated, and a route planning model is used to generate a pedestrian volume minimum route after the update is completed.

10. The improved recommendation algorithm based travel route customization method according to any one of claims 6-9, characterized in that for the route provided by the travel APP, the user can click to view the address, picture, price, poor rating and comment information of the coordinates in the route.