WO2021140371A1

WO2021140371A1 - Decision management system with hybrid strategy optimization for tourist's travel planning

Info

Publication number: WO2021140371A1
Application number: PCT/IB2020/060665
Authority: WO
Inventors: Sanam HAJSALEHI; Ehsanollah TOOKALOO; Amirabbas SAFFARI
Original assignee: Hajsalehi Sanam; Tookaloo Ehsanollah
Priority date: 2020-11-12
Filing date: 2020-11-12
Publication date: 2021-07-15

Abstract

A decision management system for travel planning of tourists comprising a traveler recommendation system based on user similarity in the database section, a hybrid recommendation system comprises of item base section and user base recommendation section and a report evaluation system wherein the reliability of the users, reports and reviews is predicted and finally system updated the reports by self-training methods for future uses.

Description

Title of Invention: Decision management system with hybrid strategy optimization for tourist's travel planning

OOOI!Technical Field

The present invention relates to a decision management system for applying strategies to manage clients, such as customers, or applicants, of a database. More specifically, the present invention relates to a decision management system having hybrid strategy optimization capabilities for tourist travel planning.

Background Art

With the popularity of Internet technology and the rise of e-commerce sites, online travel information search has become the main way for tourists to make travel plans. But with the massive increase of online travel information, tourists can no longer make decisions in the face of these explosive travel information. The travel recommendation system is a good way to solve the information overload.

Recently, industry and academia have been studying and developing tourism recommendation systems for providing convenient travel information to tourists, including next point of interest suggestion, and point of interests travel route recommendation

Summary of Invention

This system is a hybrid recommendation system that works based on both item base and user base methods. The item base method means similarities between items and options that the user has been interested in so far. The user base method is based on user behaviors and similarity of the present user with other users. As a result, the system can offer suitable options to the user by report evaluation system that evaluate reliability of users and updated previous reports.

Technical Problem

Tourism has become more and more popular with the improvement of people's living standards. In order to arrange travel itinerary before traveling, tourists often need to use the Internet to search and obtain travel information, but a large amount of travel-related information on the Internet often brings users the problem of overload information makes it impossible for users to effectively and accurately obtain the travel information they are interested in. Applying the traditional recommendation system to the tourism industry, through the establishment of a tourism recommendation system, it can effectively help tourists make travel planning, simplifying tourists' travel preparations. However, due to the particularity and complexity of tourism, traditional recommendation techniques often cannot fully meet the individual needs of tourists. In response to the above problems, this invention proposes a travel recommendation system based on a hybrid recommendation strategy: including user interest modeling and hybrid recommendation from the previous reviews

Solution to Problem

This system is a hybrid recommendation system that works based on both item base and user base methods. The item base method means similarities between items and options that the user has been interested in so far. The user base method is based on user behaviors and similarity of the present user with other users. As a result, the system can offer suitable options to the user.

For example, in this system, a user has been similar to other users who were interested in a particular food, which is why the system offers such option to the person, or in another example, the person has behaved similarly to users in system who have a high income, and as a result, he is offered expensive and luxurious options. The more similar the person is to the current user and the more points he / she gave to the place he/she liked, the higher the priority of this place and the points of that place will be updated. As a result, up to this stage of the process; there are a number of options that are rated either based on the behavioral similarity of the users or based on the user's own reviews.

Range of these scores are usually not equal and said scores need to be standardized. Therefore, in the recommendation hybrid section, in order to equalize the range of said scores, variance of the scores are harmonized. Then, according to the places in the system and the points assigned to it, the distance between these places and the current location of the user is obtained.

As a result, the effective parameter for this situation is the user's current location, so the previous score is measured and updated at this distance; in this case, the greater the distance, the lower the assigned score is. So the said score is determined based on an equation inversely of the distance to the person Then, the places are sorted according to their score, and places with higher scores are given higher priority. It then returns the suggested locations along with their points to the system so that the system can notify the user of this information.

Another point that is used to identify places is that people who are similar to this person in any way usually state where they have gone and what their opinions are about those places, which using the same data and Among these parameters, it can be said which places are suitable for a person.

In this system, some data sets are used for this purpose in the database, which can be in the form of table or any type of data set. One of these sets is each person's profile parameters, which include job, income or budget, age, place of residence, and other characteristics that can affect user tastes, such as gender, marital status, number of children, and self-reported interests.

Based on the above data, this system finds the closest profiles among registered users, using clustering algorithms or the KNN (k-nearest neighbors) algorithm or any other classification algorithm, and then it must be estimated on what basis the closest profiles are spaced apart and what is the distance between them to find a score for their similarity percentage. The greater the distance, the less likely the system is to suggest where the person has gone. Then, each of those users or profile holders will be put together with their points. Again, this is done based on another data set, which is user following graph. Here, in terms of graphical distance that users have with each other, their similarity is recognized and they are scored to be used in recognizing similarity among users. Afterward, the reviews that the users wrote about the places are obtained, for example, one of the previous users commented about a specific place and posted a number of photos and videos and rated that place. Accordingly, the system selects the most similar reviews that should be disaggregated by location, and the MSE criterion is obtained based on the maximum difference between the reviews of the two users, which states dissimilarity between two users. If necessary, the system can sort any number of users that are most similar to the main user.

There are four features for each report in report evaluation system: 1) Features related to the new report, such as type, rank and score of the place, which are possible to specify in two ways or methods:

The first method in which is determined by own user and the second method is based on the likes and dislikes that other users have reported for it.

2) Find similar reports for a place from which the weighted average is later obtained, which is based on the same rank and score of the report and the percentage guessed by the system that we guessed our model is right or wrong.

3) User reliability which is the degree of user uprightness and morality in the system that determines by the average score given by the reports.

4) The user has several followers that determine the importance of the user within the system.

These four properties are given to a classification model to determine the probability of the correctness or incorrectness of this model. To do this, a model is used that can support an online learning or a smart learning technique and it is not necessary for the system to be trained from the beginning. For example, in this case, we can use the models used in deep learning, which are very accurate for work in this field. When the probability score is calculated by this model, this score is imported into the user's credit set, now a report is added to the set and can be used for future reports. Then the score of the report is set, in this case the reliability is updated.

As shown in Fig.1 traveler recommendation system based on user similarity in the database section information is stored from four sources: users followers parameters , users following graph, users reviews and uses reports. This data is processed in the recommendation section.

In one set of the recommendation system similar profile is detected by KNN or clustering algorithm. Then the distance profile is calculated by using distance metrics. Each distance profile was scored with 1/ (Distance) parameter and each profile owner is mapped by said score.

In another set, similar users are found by similar graph and then distance of the users and others in graph is calculated. Each said distance is scored by 1/ (Graph Distance) parameter and then each users is mapped by score of the 1/ (Graph Distance) parameter.

In third and fourth sets, similar reviews or reports in accordance with place of users are found, then MSE of user rate is calculated and then scored by 1/(MSE) parameter, then each review or report is mapped.

The data of the aforementioned scores standardize in recommendation engine, and then similar user is found by clustering techniques and then data of similar users is return ordering by similarity scores.

As shown in Fig.2 the hybrid recommendation system comprises of item base section and user base recommendation section. The score of place by reviews or reports of users comes from item base recommendation section and scores of place by user similarity comes from user base recommendation section are standardized in hybrid recommendation system. The distance of user location and each said location is determined and then said standardizes scores are modified by a new parameter that is score^*(1 /(Distance)).

As shown in Fig.3 the report evaluation system has two main features. In the database and data collection section user reliability scores and user following graph and reports are stored. In second section the new report is formed by the type, place and rating of previous reports which has been updated in the prediction section. The person reliability is predicted by number of user’s followers which is obtained from user following graph and the median of scores of user report as well as weighted average report rating. In prediction sector, model classified with incremental learning support to get reliability score. The new report reliability is inserted to user reliability collection and score of report’s reliability is updated. If the reliability is less than auto ignore criterion, report statues will be changed to ignorance. If the reliability score is more than auto ignore criterion, two sub-flowcharts or two processes are created. In one process, the system alerts the admin to decide report status and in second process the report status updated to pending or auto-action done status.

The system has a control panel that can be configured by the admin with three parameters inside: If the report is less than the specified value, the said report is ignored. This action is specified inside the panel by the admin.

If the report is less than the notify criterion, the system warns the admin what to do with this report, i.e. ignore this report or apply it and, for example, deactivate one of the locations or not.

If the report score is more than the notify criteria, admin makes the decision and then the model is trained again by the admin decision. Now, if it is more than the notify value, it will be discussed in auto action whether an automatic task was defined for it or not. If it is not defined, it will wait for the report message and the admin will decide on it later. Finally, the system automatically updates the report message and saves the report for future revisions.

Brief Description of Drawings

Fig.1 shows place recommendation system based on user similarity in the database.

Fig.2 shows the hybrid recommendation system comprises of item base section and user base recommendation section

Fig.3 shows the report evaluation system in which the main feature of evaluation system and flowchart of prediction and evaluation of reports and users are displayed.

Claims

[Claim 1] |A decision management system for travel planning of tourists comprising:

A traveler recommendation system based on user similarity in the database section which its information is stored from four sources: users follows parameters, users following graph, users reviews and users reports, a hybrid recommendation system comprises of item base section and user base recommendation section, the score of place by reviews or reports of users comes from item base recommendation section and scores of place by user similarity comes from user base recommendation section are standardized in hybrid recommendation system, a report evaluation system wherein the reliability of the reports and users is predicted and new reports are created and are inserted in dataset based on the smart learning algorithm j

[Claim 2] [The decision management system of claim 1 further comprising a control panel that can be configured by the admin with three parameters inside:

A) If the report is less than the specified value, ignore the said report,

B) If the report is less than the notify criterion, it warns the admin about the future action,

C)lf the report score is more than the notify criteria, admin makes the decision and then the model is trained again by the admin decision, finally, the system automatically updates the report message and saves the report of what it has done for future revisions.

[Claim 3] [The decision management system of claim 1, wherein the recommendation system detects similar profile by a classification algorithm and then distance profile calculate by using distance metrics and each distance profile was scored with 1/ (Distance) parameter and each profile owner is mapped by said score j

[Claim 4] [The decision management system of claim 1, wherein the similar users are found by similar graph and then distance of the users and others in graph is calculated and each said distance is scored by 1/ (Graph Distance) parameter and then each users is mapped by score of the 1/ (Graph Distance) parameter i

[Claim 5] The decision management system of claiml, wherein the third and fourth sets, similar reviews or reports in accordance with place of users are found, then MSE of user rate is calculated and then scored by 1/(MSE) parameter, then each review or report is mapped. |

[Claim 6] i The decision management system of claim 1 and 3, wherein the classification algorithm is KKN (k-nearest neighbors) algorithm i