1 of 16

BUDGET

BASED

TRIP �PLANNER

2 of 16

TABLE OF CONTENT

PROJECT OVERVIEW

STEP 1

TECH STACK

STEP 2

LIMITATIONS AND

FUTURE SCOPE

STEP 5

ARCHITECTURE

STEP 3

DEMO

STEP 4

3 of 16

PROJECT OVERVIEW

Our project is a web-based application that simplifies the trip planning process and enhances the overall travel experience for users. By providing personalized recommendations for fun activities, for different time of the day based on the user's budget and preferences, as well as allowing them to fully customize their itinerary, our application aims to alleviate the stress associated with vacation planning and help users enjoy their destination to the fullest.

Project Proposed:

4 of 16

Budget Based planning.
Personalized trip activates according to users preferences and ratings.
Top recommendations for fun activities of day and night.
Customizable itinerary planning.
Weather information of different places.
Distance measurement between different places based on users choice.
Downloadable PDF itinerary.
Easy to access UI/UX.

Project Features:

5 of 16

TECH STACK

Frontend:

Backend:

Project Tracking

Tools

Deployment

6 of 16

PIPELINE

RECOMMENDATION SYSTEM

We used scraped data from tripadvisor using lxml and requests python libraries, which process the HTML files using elementTree API.HTML string into a tree-like structure

We have data for about 3500 attractions and about 30,000 reviews with user name & rating . We have used gathered data with parameters like, attraction_id, category, url, location, country, province, price,rating, user_id.

The next part is data cleaning, where we clean and integrate data from different tables to transform the data into a more usable form. We removed white spaces, hyphens, and converted the text tags to lowercase. The resulting cleaned text will be in all lowercase and have underscores instead of spaces between words. We also had missing rating, price, and coordinate values. To deal with them, initially we put -1 there. To deal with missing rating, price, and coordinate values. We calculated the average values for each combination of province and category, excluding attractions with missing ratings (-1). And later on we fill those null values with the average values. Also, Transformed the location column into 2 separate columns for latitude and longitude. We calculate the average latitude and longitude for each city and category where both latitude and longitude are not null, and store the results. SO we can say that, The missing coordinate values were handled based on the assumption that similar categories of attractions will be present in the same locality. And finally we split the dataset into 75% training and 25% validation.

The preprocessed data is then fed into a Restricted Boltzmann Machine (RBM) model, which is a type of undirected deep neural network used for collaborative filtering. The RBM model is trained on the attraction dataset. We created a rating matrix with our dataset. . The ratings provided by the user i for attraction j was filled at (i,j) position of the matrix. The positions without any user-item interaction were filled using a 0 rating. Each row of the matrix was then passed to the visible layer of the RBM model for training with weights and bias initialized randomly. The model performance was evaluated using the Contrastive Divergence algorithm and the model with the best performance, i.e., least divergence between input and reconstructed output at the visible layer was saved.

It then computes average ratings for each attraction category for each user and calculates a similarity score between the input user and other users based on the similarity of their category-rating profiles.
The most similar user is then selected and used to load a pre-trained RBM model.
The model is used to generate recommendations based on the user's past ratings, which are divided into seen and unseen attractions.

This saved model was used to obtain recommendations for all the users in the existing dataset. Then the traveler is matched with the closest user in the dataset based on attraction category preferences and the recommendations are obtained for the traveler.

A very common problem in providing recommendations to users is the cold-start problem. That is, providing recommendation to a new user who does not exist in the dataframe. So For a new user, there is no information on what he/she likes. In order to solve this problem, we need to profile the user and understand his/her preferences. So 20 Most frequent/popular categories of tours were taken and provided as options to users. Users are asked to select and rate at least five of the provided categories. This information then used to match the new user with a user in existing dataframe sharing similar preferences. The recommendation obtained for the matched user in the dataframe was provided to the new user as his attraction recommendations.

OVERALL RECOMMENDATION SYSTEM:

The function first reads in data and preprocesses it by cleaning and transforming it into a format that can be used to train the RBM.
It then computes average ratings for each attraction category for each user and calculates a similarity score between the input user and other users based on the similarity of their category-rating profiles.
The most similar user is then selected and used to load a pre-trained RBM model.
The model is used to generate recommendations based on the user's past ratings, which are divided into seen and unseen attractions.

Finally, the function exports the recommendations and the trained model to files. Load the unseen attractions from the RBM model for the specified user, and join them with the full attractions dataframe to get more information about the recommended attractions. It filters this joined dataframe by the specified price range and province, and returns the resulting data frame with URLs for each attraction.

Hence, distance between consecutive visits for attractions had to be considered. The top recommendation with highest score was chosen, distance was calculated between every attraction to our selection and the closest attraction with next highest recommendation score was chosen for next recommendation.

7 of 16

DEMO VIDEO

8 of 16

9 of 16

PROJECT ARCHITECTURE

10 of 16

CICD/Deployments

11 of 16

SPRINT JOURNEY

SPRINT 2

Model Work and Frontend Prototype

SPRINT 4

Frontend components

Project Research and Learning

SPRINT 1

Backend and CI/CD deployment scripts

SPRINT 3

Frontend Components and Deployment

SPRINT 5

12 of 16

CHALLENGES

Learning different technologies
Data Set Retrieval
Improving Model for better personalized recommendations
State management with Vuex
Showing final itinerary
Key management on EC2 machine and GitHub machine

Learning different technologies:

We wanted to expose ourselves with new technologies. So it was a challenge and time taking task, but we were able to learn it by putting continuous effort.

Data Set Retrieval: data set scraping was a not working properly from trip advisor, so we used a scraped dataset from trip advisor we found online

Improving Model for better personalised

recommendations

It was a challenge for us to give users customiable prefernces, to overcome this we implemented a rating based recommendation system.

Store management with Vuex

Since it’s a SPA all the elements that you see are individual components which use props data or use Vue store., it was a hassle to transfer data from many parent and child components through props, that’s why we used vuex state management to connect different components.

Showing final itinerary:

It was a challenging task, however we successfully implemented it

Key management on EC2 machine and GitHub

machine

Key management was a challenege as there were multiptl keys because we had different EC2 machines for deployment. To overcome it we used github secrets and deployment keys.

13 of 16

FUTURE SCOPE

Adding more categories to ML model
Finding better ways to rate the categories as compared to the current rating system
Adding more travelling instructions
Containerization of ML model as well as backend.
Adding restaurants and hotel recommendations.
Enabling the booking feature for the itenary
End to end testing

��

14 of 16

RETROSPECTION

WENT GREAT

80%

NEEDS IMPROVEMENT

20%

15 of 16

LEARNING AND CONCLUSION

Our travel planning platform is designed to make the process of planning and enjoying a trip easier and more enjoyable. With features such as budget-based planning, personalized trip activities, top recommendations for fun activities, customizable itinerary planning, weather information, distance measurement, and a user-friendly interface, our platform provides a comprehensive and seamless experience for travelers of all types.

The project also provided an opportunity to work collaboratively as a team and gain valuable knowledge and skills in web development, data analysis, and project management. Overall, the project was successful in creating a useful tool while fostering team collaboration, skill development and linking all technologies together.

16 of 16

THANK YOU!