WO2019184833A1

WO2019184833A1 - Tourism information recommending method and device

Info

Publication number: WO2019184833A1
Application number: PCT/CN2019/079335
Authority: WO
Inventors: 黄晓鸣; 陈怒潭; 顾元勋; 陈旭亮
Original assignee: 上海程向信息科技有限公司
Priority date: 2018-03-27
Filing date: 2019-03-22
Publication date: 2019-10-03
Also published as: CN108537373A

Abstract

Disclosed in the present application are a tourism information recommending method and device, The method comprises: a user information obtaining step: acquiring data of a first user, the data of the first user comprising identity information of the first user and history information of the user; and a recommendation information generating step: inputting the data of the first user into a trained deep neural network, and generating a tourism information recommendation result which expresses the probability that the first user will go to at least one tourist attraction, wherein the deep neural network is obtained by means of a training step, and the training step comprises: a user data marking step, a tourist attraction data acquisition step, a model prediction step and a model correction step. The described method employs a machine learning method to automatically recommend tourist attractions for a user by acquiring and analyzing behavior data of the user, which saves the time of the user and brings a brand new and better user experience.

Description

Tourist information recommendation method and device

Technical field

The present application relates to the field of information processing, and in particular, to a travel information recommendation method and apparatus.

Background technique

At present, the application for tourism (APP) generally adopts the following methods when recommending tourist attractions for users. One method is to require the user to select the desired destination, to recommend the attraction according to the user's choice and to make a schedule. The method requires the user to input the target attraction and calculate, when the user does not know which destination to go or the information of the attraction is not clear. This method is not applicable, so it is less practical and not suitable for the general public. There is also a method that is done manually by the professional customizer of the travel agency. First, the customizer needs to know a lot of travel information and information, and has a good understanding of the characteristics of each place. Secondly, the customizer needs to spend a lot of time communicating with the user and communicating according to the communication. As a result, the recommended attractions are adjusted. This method requires a lot of human resources and is inefficient.

Summary of the invention

The object of the present application is to overcome the above problems or at least partially solve or alleviate the above problems.

According to an aspect of the present application, a travel information recommendation method is provided, the method comprising:

a user information obtaining step: acquiring first user data, where the first user data includes identity information of the first user and user history information;

a recommendation information generating step: inputting the first user data into the trained deep neural network to generate a travel information recommendation result, the travel information recommendation result characterizing a probability that the first user will go to at least one attraction tour;

Wherein, the deep neural network is obtained through a training step, and the training steps include:

User data labeling step: for each user in the user set, acquiring the travel history data of the user, and using the data before the time point T in the travel history data as the input data of the training data, and the time point T After that, the first tourist attraction that the user has visited and other attractions outside the tourist attraction are marked separately to obtain actual tourism result data;

Sight data acquisition step: obtaining, for each attraction in the collection of attractions, attraction data, the attraction data including basic information data of the attraction and behavior data of each user for the attraction;

a model prediction step: inputting the attraction data, the identity information of each user, and the training data of the user into a deep neural network to obtain a probability set of each of the users going to each attraction tour;

The model correction step: comparing the actual travel result data with a probability set, and modifying the deep neural network to obtain the trained deep neural network.

The method adopts a machine learning method to acquire and analyze the user's behavior data through the image of the attraction and social sharing, and automatically recommend the tourist attraction for the user, and the user can obtain the result that meets the user's inner expectation without selecting the tourist attraction to be visited in advance. Saves users time and brings a new and better user experience. Social sharing, etc., plays an important role in the data analysis of this application, providing a source of data for the model and providing a good platform for capturing user behavior.

Optionally, the model prediction step includes:

Input step: inputting the training data and the attraction data to an input layer of the deep neural network;

Processing step: converting each item of the training data and each item of the attraction data into corresponding feature data;

a conversion step: converting feature data corresponding to the training data into a user feature matrix, and converting feature data corresponding to the attraction data into an attraction feature matrix;

Correlation step: associating the user feature matrix with the attraction feature matrix of each attraction, and calculating a probability that the user represented by the user feature matrix goes to the attraction;

Output step: calculating a probability set of each user in the set of users going to each attraction tour and outputting the probability set.

Through this method, a deep neural network model that is more in line with the user's habits and consistent with the user's recent behavior and psychology can be trained, so that the prediction result by the model is more accurate.

Optionally, the model correction step comprises:

a data type conversion step: comparing the probability data in the probability set with a preset threshold, and converting the probability data into an integer type of data;

a residual calculation step: comparing the integer type data with the actual travel result data to obtain residual data;

Correction step: correcting the deep neural network by reverse neural propagation using the residual data.

The method analyzes the actual behavior of the user and corrects the model as a feedback result, so that the model is more accurate, and the travel recommendation result closer to the user's real requirements and expectations is calculated.

Optionally, after the step of generating the recommendation information, the method further includes:

Sight preference step: at least one attraction included in the travel information recommendation result and behavior of the first user to the at least one attraction at every predetermined time interval or in response to the first user's route planning instruction Data identifies preferred attractions;

Route planning step: path planning is performed on the preferred attraction according to the objective function and the constraint, and a travel route is generated.

The method adds route planning steps after recommending tourist attractions, and incorporates the perspective of tourism psychology into the design of the method, so that the results are more in line with user needs.

Optionally, the method further includes: after the route planning step:

a user intention determining step of: determining whether the travel route meets the expectation of the first user, and if so, performing a route determining step, and if not, performing an attraction marking step;

a route determining step: determining whether the travel route needs to be modified, and if it is necessary to modify, re-planning the route according to the modification of the scenic spot in the travel route by the first user;

Attraction marking step: marking the preferred attraction as the behavior data of the first user according to the feedback of the first user, so as to be the data required for the training step.

Through this method, the user's real feedback data can be obtained, and the data can be used as the data of the modified model to make the model more accurate.

According to another aspect of the present application, a travel information recommendation apparatus is provided, including:

a user information obtaining module, configured to acquire first user data, where the first user data includes identity information of the first user and user history information;

a recommendation information generating module configured to input the first user data into the trained deep neural network to generate a travel information recommendation result, the travel information recommendation result characterizing a probability that the first user will travel to at least one attraction;

The deep neural network is obtained through a training module, and the training module includes:

a user data labeling module configured to acquire, for each user in the user set, travel history data of the user, and use data before the time point T in the travel history data as input data of training data, After the time point T, the first tourist attraction that the user has visited and other attractions outside the tourist attraction are respectively marked to obtain actual tourism result data;

An attraction data acquisition module configured to acquire attraction data for each attraction in the collection of attractions, the attraction data including basic information data of the attraction and behavior data of the user to the attraction;

a model prediction module configured to input the training data and the attraction data into a deep neural network to obtain a probability set for each user in the user set to travel to each attraction;

And a model correction module configured to compare the actual travel result data with a probability set, and modify the deep neural network to obtain the trained deep neural network.

The device adopts a machine learning method to automatically recommend the tourist attractions by acquiring and analyzing the user's behavior data, and the user can obtain the result that meets the user's inner expectation without having to select the tourist attraction to be visited in advance, thereby saving the user's time. Come to a new and better user experience.

Optionally, the model prediction module includes:

An input module configured to input the training data and the attraction data to an input layer of the deep neural network;

a processing module configured to convert each item of the training data and each item of the attraction data into corresponding feature data;

a conversion module configured to convert feature data corresponding to the training data into a user feature matrix, and convert feature data corresponding to the attraction data into an attraction feature matrix;

An association module, configured to associate the user feature matrix with an attraction feature matrix of each attraction, and calculate a probability that the user represented by the user feature matrix goes to the attraction;

An output module configured to calculate a probability set for each of the user sets to travel to each attraction and output the probability set.

Optionally, the device is further connected after the recommendation information generating device:

a scenic spot preference module configured to, according to the first user's route planning instruction, at least one attraction included in the travel information recommendation result and the first user pair the at least one Behavioral data of the attraction determines the preferred attraction;

A route planning module configured to perform path planning on the preferred attraction according to an objective function and a constraint to generate a travel route.

According to another aspect of the present application, a computer apparatus is provided comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program The travel information recommendation method as described above is implemented.

According to another aspect of the present application, there is provided a computer readable storage medium, preferably a non-transitory readable storage medium having stored therein a computer program, which when executed by a processor, is implemented as described above Tourist information recommendation method.

The above and other objects, advantages and features of the present application will become apparent to those skilled in the <RTI

DRAWINGS

Some specific embodiments of the present application will be described in detail below by way of example and not limitation. The same reference numbers in the drawings identify the same or similar parts. Those skilled in the art should understand that the drawings are not necessarily drawn to scale. In the figure:

1 is a flow chart of one embodiment of a travel information recommendation method according to the present application;

2 is a flow chart of one embodiment of a training step in accordance with the method of the present application;

3 is a flow diagram of one embodiment of a model prediction step of a method in accordance with the present application;

4 is a flow diagram of one embodiment of a model modification step in accordance with the method of the present application;

FIG. 5 is a flowchart of another embodiment of a travel information recommendation method according to the present application; FIG.

6 is a block diagram of one embodiment of a deep neural network in accordance with the present application;

Figure 7 is a block diagram of one embodiment of a travel information recommending apparatus according to the present application;

Figure 8 is a block diagram of one embodiment of a training module of the apparatus in accordance with the present application;

9 is a block diagram of one embodiment of a model prediction module of a device in accordance with the present application;

FIG. 10 is a block diagram of another embodiment of a travel information recommending apparatus according to the present application.

detailed description

According to an aspect of the present application, a travel information recommendation method is provided. 1 is a flow chart of one embodiment of a travel information recommendation method in accordance with the present application. The method includes:

The recommendation information generating step: inputting the first user data into the trained deep neural network to generate a travel information recommendation result, and the travel information recommendation result characterizes a probability that the first user will go to at least one attraction tour.

The first user data is used to describe basic information and historical information of the user, and the basic information may be identity information, and the basic information may include one or more of the following data: user identification information (user ID), name, gender, age, Occupation, family status, user label. User tags can include travel preferences and personal preferences such as: adventure, sports, beach, leisure, sightseeing, activities, culture, adventure, party, people, food, ocean, mountain, nature, city, museum, travel, quiet, backpack, outdoor , buildings, lakes, sunsets, forests, sunrise, warmth, cold, etc. Historical information is used to describe the behavior history of the user on the social platform. For example, the social platform may include a social networking website and/or an application (APP). The historical information may include one or more of the following data: user travel history data, user-related travel-related search records, spending habits, and attention content. In a preferred embodiment, the user travel history data can be used to describe the travel history of the user. For example, the "features" of each attraction that the user has been to. For example, a record published by a user on a social platform about which of the attractions the user visits, the data may be a picture and/or text, and the text may include a log, a message, a status of the publication, and the like. The user retrieval data may include a record of the user retrieving an attraction or information related to the attraction on the social platform. Information related to the attraction may include one or more of the following data: food around the attraction, accommodation, transportation, tickets, other attractions associated with the attraction.

Then, the first user data is input into the trained deep neural network to generate a travel information recommendation result, and the travel information recommendation result represents a probability that the first user will go to at least one attraction tour. If you get the probability that the user will go to multiple attractions, the probabilities are sorted in descending order. You can give priority to recommending and displaying the most probable attractions, as well as displaying the first few attractions.

The deep neural network model, also known as the multilayer perceptron model, is an artificial neural network with a forward structure that maps a set of input data to a set of output data. It can be thought of as a directed graph consisting of multiple node layers, each connected to the next layer. In addition to the input nodes, each node is a neuron (or processing unit) with a nonlinear activation function. The deep neural network generally includes an input layer, a hidden layer, and an output layer, wherein the hidden layer includes at least one layer, or may be two or more layers. The process of neural network is divided into forward process and reverse process. The forward process is generally used for prediction, and the reverse process is generally used for training.

In the forward process, each neuron has a weight, offset, and activation function for an input. The activation function can include one or more of identity, sigmoid, ReLU, and variations thereof. In the process of using the deep neural network for prediction, the data is input to the input layer, that is, the first layer, and the output result is obtained after the neuron operation, and then the output result of the first layer is used as the input of the second layer. And so on, until the output layer outputs the final result. If the deep neural network is trained, the weights and offsets have been determined. For a new input, the prediction process can be output through the above process. In the present application, the association relationship between the user data and the attraction data is established in the deep neural network, so the first user data is input into the deep neural network, and the travel information recommendation result can be obtained, and the result can represent the first The probability that the user will travel to at least one attraction.

In the reverse process, the deep neural network is obtained by the following training steps, and Figure 2 is a flow diagram of one embodiment of the training steps of the method according to the present application. The training steps include:

User data labeling step: for each user in the user set, acquiring the travel history data of the user, and using the data before the time point T in the travel history data as the input data of the training data, and the time point T After that, the first tourist attraction that the user has visited and other attractions outside the tourist attraction are marked separately to obtain actual tourist result data. The training data can also include basic information of the user.

The set of users may be all users who registered one or more travel-related social platforms. In an optional implementation, the social platform may be a specific platform developed according to the requirements of the present application, and the platform may be a tourism-themed platform on which users can publish travel-related articles, pictures, Comments, in addition, users can also like, click on the platform recommended connections, payments, and so on.

The user's travel history data may be travel related data that has been generated after the user operates on the platform. The travel history data is divided into two parts at a certain time point, for example, the time point T, and the data occurring before the time point T in the travel history data is extracted, and the data is used as input data of the training data. . By analyzing the travel history data between the time point T and the current time, it is possible to obtain whether the user travels to some or some attractions. If the user has visited at least one tourist attraction, the first tourist attraction that the user goes to is marked as 1 and the other tourist attractions are marked as 0. Other attractions may be those that the user has visited after going to the first field attraction. It can also be that the user has not been to. The tourist attractions that have not been visited may be a certain range of attractions around the tourist attractions. It can be understood that other identification methods can also be adopted as long as the first tourist attraction that has been visited can be distinguished from other tourist attractions. After marking each attraction, the result is actual travel result data. For a user, a number of time points T can be set. For example, T can take January 1, 2015, July 1, 2016, etc., in this way, multiple training data can be generated for the entire data of one user. The formation of multiple samples increases the amount of data in the training data set, making the training results of the deep neural network model more accurate.

The training step further includes: an attraction data acquisition step. Obtaining attraction data for each of the attractions in the collection of attractions, the attraction data including basic information data of the attraction and behavior data of each user to the attraction;

The collection of attractions can include: attractions within the country of the user, attractions and venues around the world. Among them, the venues include but are not limited to: networking parties, concert venues, fashion show venues, etc. The basic information data of the scenic spot may include one or more of the following data: attraction identification information (attraction ID), attraction type, article, picture, rating, and comment. Among them, the article includes an article describing and/or introducing the attraction, and an article describing the store and the merchant in the attraction. The article can be a log. Comments include comments on attractions, articles, pictures, and attractions. Intelligent analysis of logs and comments can be implemented using semantic analysis techniques. The basic information data of the attraction is a statistical analysis performed for all users in the user collection.

The behavior data of the user in the user collection for the attraction may also be referred to as user dependency data, which characterizes the user's dependence on the attraction. For example, for the first user, when performing the scenic spot data obtaining step, for a certain scenic spot, the behavior data includes the first user's likes of the scenic spot, browsing the articles related to the scenic spot, the duration, the number of times, and the comment. Behavioral data is a statistical analysis of the behavior of a user in a data tagging step.

The training step further includes a model prediction step. The attraction data, the identity information of each user, and the training data of the user are input into a deep neural network to obtain a probability set of each of the users going to each attraction.

3 is a flow diagram of one embodiment of a model prediction step of a method in accordance with the present application. In an optional embodiment, the model prediction step may include an input step of inputting the training data and the attraction data to an input layer of the deep neural network.

The model prediction step may further include a processing step of converting each item of the training data and each item of the attraction data into corresponding feature data. The feature data can be in the form of a vector. Each item of data has a one-to-one correspondence with the feature data. For example, converting ID information into ID features converts articles into article features. After the input layer processes the input data, the obtained feature data is transmitted to the hidden layer. For the user's travel history data and the search record related to the travel, when inputting, a plurality of records can be input, and the feature data obtained in the processing step is the average feature data of the plurality of records. Taking the user's travel history data as an example, since the amount of travel history data may be large in a certain period of time, a vector can be obtained when characterizing each travel history data, due to the limitation of the model on the data input format. When performing model training, the average feature data obtained by averaging the feature data corresponding to each of the plurality of travel history data of one user is substituted into the deep neural network for calculation. The “user travel history information feature” is obtained by summing the travel history data and then taking the mean value. The obtained results can meet the requirements of the model for the data format and also reflect all the travel history.

The model prediction step may further include a converting step of converting feature data corresponding to the training data into a user feature matrix, and converting feature data corresponding to the attraction data into an attraction feature matrix. The conversion step can be performed in the hidden layer, and the hidden layer further calculates and processes the feature data to obtain a feature matrix, and the hidden layer used in the calculation can be one layer or multiple layers. Alternatively, all of the feature data associated with the user can be transformed into a user feature matrix, and all feature data associated with the attraction can be transformed into an attraction feature matrix.

The model prediction step may further include an associating step of associating the user feature matrix with an attraction feature matrix of each attraction, and calculating a probability that the user represented by the user feature matrix goes to the attraction. The association step can also be done at the hidden layer.

The model prediction step may further include an outputting step of calculating a probability set for each of the user sets to go to each attraction tour and outputting the probability set. This step can be performed at the output layer. During the training process, the output layer of the deep neural network outputs a probability that characterizes the likelihood that the user will travel to the attraction.

6 is a block diagram of one embodiment of a deep neural network according to the present application. Optionally, the deep neural network model includes three layers, in which the lowest layer represents the input layer, the penultimate layer is the hidden layer, and the reciprocal The third layer is the output layer. Part of the data in the input layer is connected to the data processed in the hidden layer through the fully connected layer, for example, user ID, gender, occupation, age, travel preferences, user's travel history data, user search history, and attraction data in the training data. The attraction ID, the type of attraction, the name of the attraction, the rating, the comment, the like, the first user's likes of the attraction, the duration and number of times the picture or log is viewed. For the logs in the attraction data, a convolutional neural network (CNN) connection is used between the input layer and the hidden layer. For pictures in the attraction data, the input layer and the hidden layer are previously connected using a recurrent neural network (RNN). In the figure, "user attraction relevance" is a probability value. The loss function of the training model is cross entropy. After inputting N scenic spots information, N "user attraction relevance features" probability values are obtained, and sorting is performed according to the probability values. The softmax activation function is implemented between the user feature and the user's scenic spot correlation feature. The function can be applied to different kinds of neural networks such as multi-layer neural network and convolutional neural network.

The present application extracts features by deep learning, for example, using a deep neural network model to calculate the relevance of a user to an attraction. In the hidden layer of the model, the data with similar features will gradually approach, and finally the feature matrix is obtained. The output layer can sort the attractions and/or activities and recommend them to the user.

The training step further includes a model correction step. 4 is a flow diagram of one embodiment of a model modification step in accordance with the method of the present application. In the correcting step, the actual travel result data and the probability set are compared, and the deep neural network is corrected to obtain the trained deep neural network.

The model modification step includes a data type conversion step of comparing the probability data in the probability set with a preset threshold, and converting the probability data into an integer type of data. The purpose of this step is to align the probability data with the type of actual travel result data.

The model correction step further includes a residual calculation step of comparing the integer type of data with the actual travel result data to obtain residual data. The probability data can be converted to an integer and then compared with the actual travel result to obtain residual data.

The model modification step further includes a correction step of correcting the deep neural network by reverse neural propagation using the residual data. The correction method may include: an outlier test, a homogeneity test of the variance, a normality test of the error, a correlation test, and a concomitant variance stabilization transformation, a normalization transformation, and the like.

Through this method, the user does not need to input the target spot that he wants to go, and the special planner does not consult the user, and the user's behavior can be analyzed to automatically analyze the tourist attraction that the customer wants to go. This method takes into account the knowledge of tourism psychology in design, so it is more intelligent and user-friendly, and the recommendation results are more accurate.

Referring to FIG. 1 , optionally, the method may further include an attraction optimization step and a route planning step after the recommendation information generating step. The scenic spot optimization step includes: at least one attraction included in the travel information recommendation result and the first user to the at least one attraction at every predetermined time interval or in response to the first user's route planning instruction Behavioral data identifies preferred attractions. The route planning step performs path planning on the preferred attraction according to the objective function and the constraint condition to generate a travel route.

There are two ways to make a path plan after recommending a tourist attraction step. The first is to adopt a method of regularly recommending tourist routes. For example, every fixed time interval, for example, one week or one month, according to the travel information recommendation result, path planning is performed according to one or several tourist attractions. When planning with multiple attractions, these attractions preferably belong to one city or are relatively close. For example, the number of attractions, the name of the attraction, and the route planning are selected for the user according to the length of the weekend, the short holiday, and the long holiday. With this method, user time can be saved, and a practical travel plan can be provided for the user.

In another alternative embodiment, a method of regionalized path recommendation may be employed. In this embodiment, the user may select at least one city or a range of areas, and then the system selects attractions belonging to the city or the area from the recommended tourist attractions for path planning. With this method, the travel route developed is more in line with the user's psychological expectations.

Optionally, the method may also directly perform a route planning step in the recommendation information generating step. The route planning step includes: performing path planning on all or some of the scenic spots included in the travel information recommendation result according to the objective function and the constraint condition, and generating a travel route.

Using this method, intelligent analysis can be carried out according to the user's behavior data to obtain the probability of tourist attractions, and path planning is carried out according to the probability of tourist attractions. This method combines user behavior with tourism psychology, which is more in line with user needs and psychological expectations.

In an alternative embodiment, the objective function of the path plan can be as shown in equation (1):

Where R is the path with the smallest overall distance planned, c _ij is the distance from attraction i to attraction j, and x _ij indicates whether there is a path from attraction i to attraction j. If x _ij =1, there is communication between the two attractions. The path, if x _ij =0, means that there is no path between the two attractions, and n represents the total number of attractions.

The constraints of the objective function include:

0≤x _ij ≤1, i=1,...,n,j=1,...,n (2)

Among them, the formula (3) indicates that only one of the other attractions reaches the attraction j, and the formula (4) indicates that only one of the other attractions can be reached from the attraction i.

Optionally, the constraint of the objective function may further include at least one of the following conditions:

(1) Attraction opening time, the data can be obtained from articles, comments, other websites or channels published by the user on the social platform of the application;

(2) The time required to visit the attraction, which may be the time required to normally visit an attraction or calculate the average time of time for multiple users to browse the attraction under statistical conditions;

(3) Vehicles, the constraint condition may be determined according to the user's selection, or may be determined according to analysis of other users, or the user's travel travel habits may be obtained according to the user's travel history data, search records, and the like;

(4) The maximum carrying capacity of the user, and the maximum carrying capacity indicates the maximum duration of the user's daily outgoing time. The data can be learned and predicted from the user's historical travel preferences through machine learning calculations, or can be obtained through travel arrangements, such as flight information. For example, if the user is analyzed as a leisurely type through machine learning calculation, it indicates that the maximum load capacity is too small, and the user can be arranged with fewer attractions every day; if the user is a card type, the maximum load is biased. Large, you can arrange more attractions for the user every day.

(5) The starting point and the ending point of the itinerary, it can be understood that the starting point and the ending point can be the same or different, and the data can be obtained by user definition;

(6) The total length of one day or the total length of travel time does not exceed the first threshold. The total length of the journey or time does not include the distance and time of the visit. The first threshold may be obtained by machine learning predicting user historical travel data, or obtained by user customization.

(7) The total amount of time spent visiting the attraction one day does not exceed the second threshold. The total amount of time spent visiting the attraction does not include the time spent on the journey. The second threshold may be predicted by machine learning to obtain historical user travel data, or user-defined acquisition. It can be understood that the sixth point and the seventh point can also be combined to form a constraint condition.

(8) Process relevance. Specifically, for example, if the user requests that the attraction A must be visited before the attraction B, the system judges the restriction condition. For example, the priority of the attraction is determined according to the application scenario. The application scenario can be time. If time is limited, then attraction A has a higher priority than attraction B than attraction B, and attraction B may not be in the path planning result. Priority data can be obtained by sorting the attractions in the travel information recommendation results.

FIG. 5 is a flow chart of another embodiment of a travel information recommendation method according to the present application. In this embodiment, the travel information recommendation method may further include: after the route planning step:

According to another aspect of the present application, there is also provided a travel information recommending apparatus, and Fig. 7 is a block diagram of an embodiment of a travel information recommending apparatus according to the present application. The device includes:

8 is a block diagram of one embodiment of a training device of a device in accordance with the present application. In this embodiment, the model correction module includes:

a data type conversion module configured to compare the probability data in the probability set with a preset threshold, and convert the probability data into an integer type of data;

a residual calculation module configured to compare the integer type of data with the actual travel result data to obtain residual data;

A correction module configured to modify the deep neural network with the residual data.

9 is a block diagram of one embodiment of a model training device of a device in accordance with the present application. Optionally, the model prediction module includes:

Referring to FIG. 7, optionally, the device is further connected after the recommendation information generating device:

FIG. 10 is a block diagram of another embodiment of a travel information recommending apparatus according to the present application. Optionally, in the device, the route planning module is further connected with:

a user intention judging module, judging whether the travel route meets the expectation of the first user, if yes, executing a line determination module, and if not, executing an attraction labeling module;

a route determining module configured to determine whether the travel route needs to be modified, and if necessary, re-planning the path according to the modification of the scenic spot in the travel route by the first user;

An attraction tag module configured to mark the preferred attraction as behavior data of the first user based on feedback from the first user to serve as data required for the training step.

The application adopts artificial intelligence instead of human to make decision, first obtains data information through a specific channel, and finally realizes a recommendation function of fully generating a personalized travel route conforming to the user image.

The present application acquires a user's travel behavior characteristics through a social platform, and the information is obtained from the user's travel behavior and social behavior. This application mainly uses artificial intelligence to make predictions and recommendations, and uses a linear programming algorithm to perform path planning. In the recommendation part of the attraction, this application uses the basic theory of tourism psychology to verify and explain the characteristics of the user's travel behavior, and uses the machine learning algorithm to predict and recommend the user's travel behavior. By using the method and device of the invention, the user's time can be saved, and the attraction information and the route information can be automatically recommended for the user according to the behavior of the user, which is more convenient, quick and user-friendly. In the aspect of path planning, the present application formulates a travel route by means of an integer linear programming algorithm for the manner in which a user travels during a trip. Since artificial intelligence does not have a very strong advantage in accurate calculation, this application combines a linear programming algorithm to calculate the part that requires precise calculation, thus realizing the effect of "customization". The main parameters considered by the algorithm include: the point of interest selected by the user at the user terminal or the point of view of the user's optimal choice predicted by artificial intelligence, the type of the tool to be selected, the point of arrival, the path most likely to be selected by the user. , the user's time requirements for viewing attractions, and so on. The path is planned using the objective function and constraints, and the simulation route is designed. Through the design of the platform, the corresponding user's feedback and requirements are obtained, for example, whether the route needs to be modified, whether the time schedule is reasonable, whether the user is willing to purchase the scenic spot ticket or the transportation ticket on the route, and refer to the route to travel, etc., preliminary determination result. Since the existing technology only adopts the linear programming method in path planning, the application combines artificial intelligence and path planning, so that more accurate calculation results can be obtained to meet the needs of different types of customers.

Through the mobile Internet application in this platform, a large amount of user feedback can be obtained, and then the feature data is modified to optimize the algorithm parameters. Once the user determines the final route and travels with reference to the route, the platform will also be able to collect data through big data, such as user ratings of attractions, restaurants recommended by other users, more accurate traffic information, and the like.

According to another aspect of the present application, there is also provided a computer apparatus comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer At the time of the program, any one of the travel information recommendation methods described above is implemented.

According to another aspect of the present application, there is also provided a computer readable storage medium, preferably a non-volatile readable storage medium having stored therein a computer program that, when executed by a processor, implements Any one of the recommended travel information methods.

According to another aspect of the present application, a computer program product comprising instructions is also provided. When the computer program product is run on a computer, the computer is caused to execute any one of the above-described travel information recommendation methods.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer loads and executes the computer program instructions, the processes or functions described in accordance with embodiments of the present application are generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be from a website site, computer, server or data center Transfer to another website site, computer, server, or data center by wire (eg, coaxial cable, fiber optic, digital subscriber line (DSL), or wireless (eg, infrared, wireless, microwave, etc.). The computer readable storage medium can be any available media that can be accessed by a computer or a data storage device such as a server, data center, or the like that includes one or more available media. The usable medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

A person skilled in the art should further appreciate that the elements and algorithm steps of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both, in order to clearly illustrate hardware and software. Interchangeability, the composition and steps of the various examples have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the solution. A person skilled in the art can use different methods to implement the described functions for each particular application, but such implementation should not be considered to be beyond the scope of the present application.

It will be understood by those skilled in the art that all or part of the steps of implementing the above embodiments may be performed by a program, and the program may be stored in a computer readable storage medium, which is non-transitory ( English: non-transitory) media, such as random access memory, read-only memory, flash memory, hard disk, solid state disk, magnetic tape (English: magnetic tape), floppy disk (English: floppy disk), CD (English: optical disc) And any combination thereof.

The above description is only a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto, and any person skilled in the art can easily think of changes or within the technical scope disclosed in the present application. Replacement should be covered by the scope of this application. Therefore, the scope of protection of the present application should be determined by the scope of protection of the claims.

Claims

A method for recommending tourist information, the method comprising:

a user information obtaining step: acquiring first user data, where the first user data includes identity information of the first user and user history information;

a recommendation information generating step: inputting the first user data into the trained deep neural network to generate a travel information recommendation result, the travel information recommendation result characterizing a probability that the first user will go to at least one attraction tour;

Wherein, the deep neural network is obtained through a training step, and the training steps include:

User data labeling step: for each user in the user set, acquiring the travel history data of the user, and using the data before the time point T in the travel history data as the input data of the training data, and the time point T After that, the first tourist attraction that the user has visited and other attractions outside the tourist attraction are marked separately to obtain actual tourism result data;

Sight data acquisition step: obtaining, for each attraction in the collection of attractions, attraction data, the attraction data including basic information data of the attraction and behavior data of each user for the attraction;

a model prediction step: inputting the attraction data, the identity information of each user, and the training data of the user into a deep neural network to obtain a probability set of each of the users going to each attraction tour; and

The model correction step: comparing the actual travel result data with a probability set, and modifying the deep neural network to obtain the trained deep neural network.
The method of claim 1 wherein said model prediction step comprises:

Input step: inputting the training data and the attraction data to an input layer of the deep neural network;

Processing step: converting each item of the training data and each item of the attraction data into corresponding feature data;

a conversion step: converting feature data corresponding to the training data into a user feature matrix, and converting feature data corresponding to the attraction data into an attraction feature matrix;

Correlation step: associating the user feature matrix with an attraction feature matrix of each attraction, and calculating a probability that the user represented by the user feature matrix goes to the attraction; and

Output step: calculating a probability set of each user in the set of users going to each attraction tour and outputting the probability set.
The method of claim 1 wherein said model modifying step comprises:

a data type conversion step: comparing the probability data in the probability set with a preset threshold, and converting the probability data into an integer type of data;

a residual calculation step: comparing the integer type of data with the actual travel result data to obtain residual data; and

Correction step: correcting the deep neural network by reverse neural propagation using the residual data.
The method according to any one of claims 1 to 3, further comprising: after the step of generating the recommendation information, the method further comprises:

Sight preference step: at least one attraction included in the travel information recommendation result and behavior of the first user to the at least one attraction at every predetermined time interval or in response to the first user's route planning instruction Data identifies preferred attractions; and

Route planning step: path planning is performed on the preferred attraction according to the objective function and the constraint, and a travel route is generated.
The method according to claim 4, further comprising: after the route planning step:

a user intention determining step of: determining whether the travel route meets the expectation of the first user, and if so, performing a route determining step, and if not, performing an attraction marking step;

a route determining step: determining whether the travel route needs to be modified, and if it is necessary to modify, re-planning the route according to the modification of the scenic spot in the travel route by the first user;

Attraction marking step: marking the preferred attraction as the behavior data of the first user according to the feedback of the first user, so as to be the data required for the training step.
A tourist information recommendation device, comprising:

a user information obtaining module configured to acquire first user data, where the first user data includes identity information of the first user and user history information;

a recommendation information generating module configured to input the first user data into the trained deep neural network to generate a travel information recommendation result, the travel information recommendation result characterizing a probability that the first user will travel to at least one attraction;

The deep neural network is obtained through a training module, and the training module includes:

a user data labeling module configured to acquire, for each user in the user set, travel history data of the user, and use data before the time point T in the travel history data as input data of training data, After the time point T, the first tourist attraction that the user has visited and other attractions outside the tourist attraction are respectively marked to obtain actual tourism result data;

An attraction data acquisition module configured to acquire attraction data for each attraction in the collection of attractions, the attraction data including basic information data of the attraction and behavior data of the user to the attraction;

a model prediction module configured to input the training data and the attraction data into a deep neural network to obtain a probability set for each user in the user set to travel to each attraction; and

And a model correction module configured to compare the actual travel result data with a probability set, and modify the deep neural network to obtain the trained deep neural network.
The apparatus according to claim 6, wherein the model prediction module comprises:

An input module configured to input the training data and the attraction data to an input layer of the deep neural network;

a processing module configured to convert each item of the training data and each item of the attraction data into corresponding feature data;

a conversion module configured to convert feature data corresponding to the training data into a user feature matrix, and convert feature data corresponding to the attraction data into an attraction feature matrix;

An association module configured to associate the user feature matrix with an attraction feature matrix of each attraction, and calculate a probability that the user represented by the user feature matrix goes to the attraction; and

An output module configured to calculate a probability set for each of the user sets to travel to each attraction and output the probability set.
The device according to claim 6 or 7, wherein the device is further connected after the recommendation information generating device:

a scenic spot preference module configured to, according to the first user's route planning instruction, at least one attraction included in the travel information recommendation result and the first user pair the at least one The behavioral data of the attraction determines the preferred attraction; and

A route planning module configured to perform path planning on the preferred attraction according to an objective function and a constraint to generate a travel route.
A computer device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program as claimed in any one of claims 1 to 5 One of the methods described.
A computer readable storage medium, preferably a non-volatile readable storage medium, having stored therein a computer program, when executed by a processor, implements the method of any one of claims 1 to 5 method.