Personalized Travel Sequence Recommendation on MultiSource Big Social Media
Abstract
Personalized Travel Sequence Recommendation on MultiSource Big Social Media management report in data mining Big data increasingly benefit both research and industrial area such as health care, finance service and commercial recommendation. This paper presents a personalized travel sequence recommendation from both travelogues and community-contributed photos and the heterogeneous metadata (e.g., tags, geo-location, and date taken) associated with these photos. Unlike most existing travel recommendation approaches, our approach is not only personalized to user’s travel interest but also able to recommend a travel sequence rather than individual Points of Interest (POIs). Topical package space including representative tags, the distributions of cost, visiting time and visiting season of each topic, is mined to bridge the vocabulary gap between user travel preference and travel routes.
We take advantage of the complementary of two kinds of social media: travelogue and community-contributed photos. We map both user’s and routes’ textual descriptions to the topical package space to get user topical package model and route topical package model (i.e., topical interest, cost, time and season). To recommend personalized POI sequence, first, famous routes are ranked according to the similarity between user package and route package. Then top ranked routes are further optimized by social similar users’ travel records. Representative images with viewpoint and seasonal diversity of POIs are shown to offer a more comprehensive impression. We evaluate our recommendation system on a collection of 7 million Flickr images uploaded by 7,387 users and 24,008 travelogues covering 864 travel POIs in 9 famous cities, and show its effectiveness. We also contribute a new dataset with more than 200K photos with heterogeneous metadata in 9 famous cities.
INTRODUCTION
Automatic travel recommendation is an important problem in both research and industry. Big media especially the flourish of social media (e.g. Facebook, Flick, Twitter etc.) offers great opportunities to address many challenging problems, for instance, GPS estimation and travel recommendation. Travelogue websites (e.g., www.igougo.com) offer rich descriptions about landmarks and traveling experience written by users. Furthermore, community-contributed photos with metadata (e.g., tags, date taken, latitude etc.) on social media record users’ daily life and travel experience. These data are not only useful for reliable POIs (points of interest) , travel routes but give an opportunity to recommend personalized travel POIs and routes based on user ’s interest. There are two main challenges for automatic travel recommendation. First, the recommended POIs should be personalized to user interest since different users may prefer different types of POIs.
Take New York City as an example. Some people may prefer cultural places like the Metropolitan Museum, while others may prefer the cityscape like the Central Park. Besides travel topical interest, other attributes including consumption capability (i.e., luxury, economy), preferred visiting season (i.e., summer, autumn) and preferred visiting time (i.e., morning, night) may also be helpful to provide personalized travel recommendation. Second, it is important to recommend a sequential travel route (i.e., a sequence of POIs) rather than individual POI. It is far more difficult and time consuming for users to plan travel sequence than individual POIs. Because the relation- ship between the locations and opening time of different POIs should be considered. For example, it may still not be a good recommendation if all the POIs recommended for one day are in four corners of the city, even though the user may be interested in all the individual POIs.
Our work is a personalized travel recommendation rather than a general recommendation.
- We automatically mine user ’s travel interest from user- contributed photo collections including consumption capability, preferred time and season which is important to route planning and difficult to get directly.
- We recommend personalized POI sequence rather than individual travel POIs. Famous routes are ranked according to the similarity between user package and route package, and top ranked famous routes are further optimized according to social similar users’ travel records.
- We propose Topical Package Model (TPM) method to learn users and route’s travel attributes. It bridges the gap of user interest and routes attributes. We take advantage of the complementary of two big social media to construct topical package space.
OVERVIEW OF THE PROJECT
Recommendation systems and adaptive systems have been introduced in travel applications to support the travelers in their decision-making processes. Large amount of data can be collected from the Internet and travel guides, but these resources normally recommend familiar personalized Point of Interest (POI). This approach is able to recommend a travel sequence rather than personalized Points of Interest (POIs).had been already developed by many travel agencies. These recommendations focused on query conditions. However, such recommendation result usually become involve in package tourism advertisements and lack of flexibility and such recommender mechanism could not replicate important word-of-mouse effect about traveling experience. So this concluded that the recommender mechanism should be revised for TSA problem solving. Hence, this research proposed an Intelligence traveling recommender (ITR) system based on commonsense reasoning (CR) algorithm. Intelligence travelling recommender system included two reasoning processes, the first was general reasoning and the second was exception one.
Social Based Recommender System
A Tourist-Area-Season Topic (TAST) model was developed, which represents travel packages and tourists by different topic distributions, where the topic extraction was conditioned on both the tourists and the intrinsic features like locations, travel seasons of the landscapes. Furthermore, the TAST model was extended to the TouristRelation-Area- Season Topic (TRAST) model for capturing the latent relationships among the tourists in each travel group. Finally, the TAST model, the TRAST model, and the cocktail recommendation approach were evaluated on the real-world travel package data. The cocktail approach was much more effective because experimental results show that the TAST model could effectively capture the unique characteristics of the travel data. Also TRAST model could be used as an effective assessment for travel group information, by considering tourist relationships.
Based on Location
Based Social Network People share their locations on location based social networks and write their likings and disliking about those places there. By these data i.e. crowd source digital footprints, one could guess user preference to locations. A prototype system was developed which obtained users travel demands from mobile client and thus generated travel package containing multiple points of interest and their visiting sequence. This approach dissipated and improvement in accuracy and diversity according to the experimental results.
System Configuration:
H/W System Configuration:-
Processor : Pentium IV
Speed : 1 Ghz
RAM : 512 MB (min)
Hard Disk : 20GB
Keyboard : Standard Keyboard
Mouse : Two or Three Button Mouse
Monitor : LCD/LED Monitor
S/W System Configuration:-
Operating System : Windows XP/7
Programming Language : Java/J2EE
Software Version : JDK 1.7 or above
Database : MYSQL
OBJECTIVES OF THE PROJECT
The main objective of this project is to provide both personalized and sequential travel route for the users based on their POI’s. The main contributions are:
- To give a personalized travel recommendation rather than a general recommendation.
- Automatically mine user’s travel interest from user contributed photo collections including consumption capability, preferred time and season which is important to route planning and difficult to get directly.
- Ranking is performed based on the similarity between user package and route package, and top ranked famous routes are further optimized according to social similar users’ travel records.
- Author topic matrix modeling algorithm (ATMMA) is used to learn user’s and route’s travel attributes.
- It bridges the gap of user interest and routes attributes. Takes advantage of the complementary of two big social media to construct topical package space.
SYSTEM OVERVIEW
The system we proposed is a personalized POI sequence recommendation system which could automatically mine user ’s travel attributes such as topical interest, consumption capability and preferred time and season. In this section, we briefly introduce the terms used in this paper: topical package space, user package and route package. Secondly, we provide the system overview. Topic package space is a kind of space in which the four travel distributions of each topic are described by
- representative tags mined from travelogues which describe POIs within the same topic;
- the average consumer expenditure of the POIs within this topic, which are also mined from travelogues;
- distribution of the visiting season of the twelve months mined by the “date taken” attached with the community-contributed photos;
- distribution of visiting time during the day from travelogues. The usage of topic package space is to bridge the gap between user interest and the attribute of routes, since it is difficult to directly measure the similarity between user and travel sequence.
From mapping both user information and route information to the same space, we get the quantitative standard to measure the similarity of user and routes.
EXISTING SYSTEM
Mainly introduce three aspects of related works
- travel recommendation on various big social media;
- personalized travel recommendation;
- travel sequence and travel package recommendation
Point of Interest Recommendation using Author Topic Collaborative Filtering (ATCF) In this approach, a study of latest POI recommendation drawback to predict the users’ current cities is to be suggested. The challenge is tough to learn the user’s ordered information and provide personalized recommendation model. This system collects the knowledge of the author and therefore the cities. Through ATM, both the category and the user’s travel preferences are mined by modifying the latent model simultaneously. The ATM chiefly consists of two steps such as probabilistic generative model and Bayesian estimation model. Through ATM, the probabilities of every word to different topics are determined.
Drawbacks
- The existing studies related to travel sequence recommendation did not well consider the popularity and personalization of travel routes at the same time.
- It is far more difficult and time consuming for users to plan travel sequence than individual POIs.
- However, general travel route planning cannot well meet users’ personal requirements. Existing studies focused more on famous route mining but without automatically mining user travel interest.
PROPOSED SYSTEM
Automatic travel recommendation is an important problem in both research and industry. Big media, especially the flourish of social media offers great opportunities to address many challenging problems, for instance, GPS estimation and travel recommendation. Travelogue websites offer rich descriptions about landmarks and traveling experience written by users. Furthermore, community-contributed photos with metadata like tags, date taken, latitude etc. on social media record users daily life and travel experience. These data are not only useful for reliable POIs (points of interest) mining, travel routes mining, but give an opportunity to recommend personalized travel POIs and routes based on user’s interest. In offline module, the topical package space is mined from social media combining travelogues and community contributed photos.
Four travel distributions (i.e., topical interest, time, season and cost) of each topic are described in topical package space. It take the advantage of the complementation of the two social media. For example, the “date taken” of Flickr may be error with the influence of time difference. Sometimes observe in community- contributed photo the “date taken” of night scene is daytime. But the time descriptions of POIs of travelogues do not have time difference problem. In offline module, mine POIs and famous routes from community contributed photos, and obtain routes’ packages through mapping travelogues, which are related to these routes, to the topical package space. Online module focuses on mining user package and recommending personalized POI sequence based on user package. First, tags of user’s photo set are mapped to topical package space to get user’s topical interest distribution. It is difficult to get user’s consumption capability directly from the textual descriptions of photos. But the topics user interested in could somehow reflect these attributes.
For example, if a user usually takes part in luxurious activities like Golf and Spas, he is more likely to be rich. Combine user topical interest and the cost, time, season distribution of each topic to mine user’s consumption capability, preferred visiting time and season. After user package mining, rank famous routes through measuring user package and routes package. At last, optimize the top ranked routes through social similar users’ travel records in this city. Social similar users are measured by the similarity of user packages.
Advantages
- The system automatically mines user’s and routes’ travel topical preferences including the topical interest, cost, time and season.
- Recommends not only POIs but also travel sequence, considering both the popularity and user’s travel preferences at the same time.
Conclusion and further work
In this paper, we proposed a Personalized Travel Sequence Recommendation on MultiSource Big Social Media system by learning topical package model from big multi-source social media: travelogues and community-contributed photos. The advantages of our work are
- the system automatically mined user’s and routes’ travel topical preferences including the topical interest, cost, time and season,
- we recommended not only POIs but also travel sequence, considering both the popularity and user’s travel preferences at the same time.
We mined and ranked famous routes based on the similarity between user package and route package. And then optimized the top ranked famous routes according to social similar users’ travel records. However, there are still some limitations of the current system. Firstly, the visiting time of POI mainly presented the open time through travelogues, and it was hard to get more precise distributions of visiting time only through travelogues. Secondly, the current system only focused on POI sequence recommendation and did not include transportation and hotel information, which may further provide convenience for travel planning. In the future, we plan to enlarge the dataset, and thus we could do the recommendation for some non-famous cities. We plan to utilize more kinds of social media (e.g., check-in data, transportation data, weather forecast etc.) to provide more precise distributions of visiting time of POIs and the contextaware recommendation.
