Recommedation System for On-line traveling

in ai •  7 years ago 

Travel is an important part of life and has become one of the biggest industry with high growth rate. It reached 8 trillion USD revenue worldwide in 2017 having doubled since 2005. Digital travel sales reached 564 USD revenue in 2017 and its portion in the whole travel industry is getting bigger than ever before.

Travellers should collect information and arrange all the travel details including accommodation, transportation, dining and travel activities. Travellers may resort to travel agencies, but many want to solve with their own hands. Those who do not want to spend money on travel agencies should have to spend time and effort to gather information and cater for themselves.

Travel schedule recommendation system is another good option for the travellers to save time and effort. There are several approaches to provide travel package recommendation. Multi-agent based recommendation system, collaborative filtering, content aware recommendation system, based on location social network system, and based on mobile crowd source data are travel recommendation systems we know. Chen-Shu Wang and his coworkers` work is also worth reading. Even though suggested ideas are generally excellent and creative, none of them has been verified in the real world. In the real world, most travel agencies recommend among their travel packages which lacks flexibility.

The aim of this project is build a travel recommendation system which can be practically used in online travel sales. For this purpose, we need to consider combinations of several algorithms together, while other practical requirement such as response time and data collecting cost meet the standard of industry.

To consider time, recursive neural network is a strong candidate to yield time-considered personalized prediction. Strictly speaking, RNN is a good option for analyzing sequence of time. Start time and duration is also important for scheduling which will not be directly derived simple RNN model, so we plan to adopt combined or hybrid neural network model, in which weights should be carefully managed.

Cooperation with the industry has many merits, among which the data access would be crucial for the project. Gathering large amount of data is costly task, and not easily attainable in academic field. But in industry running a travel social media app, it is relatively easy to collect user activities in low cost. Sometimes users voluntarily creates contents containing such information. Measuring how good the predictions serve the users` travel needs is also a strong point of industry, and it goes without saying that industry is always keen to practical requirements such as response time, cost and user experience.

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!