London Accommodations Search-Engine

Installation

Download the repository:

git clone https://github.com/albertonuzzaci/ProgettoGestione.git

In the main directory uncompress the dataset in order to have:

mainDir/dataset/json/
mainDir/dataset/information.json
mainDir/dataset/reviews.pickle

It is advised to download used libraries in a virtual environment.

python -m venv venv
source activate

It may change depending on the operating system.

Install dependecies (be aware to be in the virtual environment directory):

pip install -r requirements.txt

Usage

Run the search engine

python main.py [-h] [--build-index] [-B | -S | -RS | -D2V]

Options:

• -h, --help show an help message and exit.
• --build-index build again the index.
• [-B | -S | -RS | -D2V] choose the weighting model to run.
  • -B run the search engine using the BM25F weighting model. This model is the default used in whoosh. Without any option it will be taken as the default model.
  • -S run the search engine using the SentimentModelWA weighting model. During the search a user can sort the results based on the feelings searched for.
  • -RSrun the search engine using the SentimentModelARWA weighting model. During the search a user can sort the results based on the feelings searched for and at the same time on the amount of reviews.
  • -D2V run the search engine using the Doc2VecModel weighting model. It use doc2vec which is a technique for generating vector representations of documents, allowing for efficient comparison and analysis of their semantic meanings.

Running requires

• Index: highly optimized data structure that contains indexed documents and their associated information, enabling fast and efficient text search and data analysis.
• Reviews file: file used in model that enable ranking by sentiment.
• Doc2Vec model and vector in a json file: file used by Doc2VecModel.

If It's the first time you run this project, all the needed file will be generated. However, this could take a long time. Therefore, a pre-built version of every file is provided so, if everything downloaded correctly, the project should starts istantly.

Please note that an error related to the searcher object may occur. In that case try to build again index running:

python main.py [-B | -S | -RS | -D2V] --build-index

Project Structure

• assets: directory containing images for graphic interface.
• dataset: the dataset.
• doc2vec: this package contains the implementation of Doc2VecModel
• evaluation: this package contains functions and queries for benchmarks.ipynb.
• gui: this package contains the GUI code.
• indexdir: directory containing the index.
• sentiment: this package contains the implementation of SentimentModelARWA and SentimentModelWA
• benchmarks.ipynb: notebook to run benchmarks
• controller.py: file serves as the central hub for interacting with and orchestrating the functionalities of various other files within a software project.
• index.py: this file builds setup the schema and then build the index which is stored in the indexdir directory.
• main.py: run the program.
• model.py: carry out the actual search function after having appropriately selected the model.
• config.py: stores configuration settings and parameters for the search-engine.

Query Language

All the version of the search-engine support Natural Query Language. Additionaly, when SentimentModelARWA rather than SentimentModelWA are launched a list of convenient Sentiments checkboxes will appear. Through this series of checkboxes an user can rank retrived documents based on Sentiments selected.

Furthermore, all the versions of the search-engine provide the following filters to the user in order to better personalize the search experience:

• #NResult using this button It's possible to decrease and increase the number of the retrieved documents.
• People using this button It's possible to decrease and increase the number of people the accommodation must be able to accommodate.
• Price Max through this scrollbar user can require accomodations specifying a maximum price
• Score Min through this scrollbar user can require accomodations specifying a minimum score (not related to sentiment)
• Neighborhood thorugh this series of checkboxes users can specify in which neighborhood is looking for an accomodation.
• Beds and Baths both the buttons allow users to specify how many beds and baths they need.

Moreover, It's also present slightly above the search-box a "did you mean ... ?" button in order to suggest to users a possible correction about the formulated query. By a simple click on text in the search-box will be changed.

GUI

The GUI is the front-end of the search engine. The main views present follow.

This is the main view for the models Doc2VecModel and BM25F (without sentiments).

Instead, this is the main view for the models SentimentModelARWA and SentimentModelWA (with sentiments). It's possible to see also did you mean '...'? tool.

When an accomodation is clicked this will be the view shown.

Dataset

• id: The unique identifier for the listing.
• listing_url: The URL link to the Airbnb listing page.
• name: The title of the listing.
• description: A detailed description of the property, including amenities and location information.
• host_name: The name of the host offering the property.
• host_id: The unique identifier for the host.
• host_url: The URL link to the host's Airbnb profile page.
• host_picture_url: The URL link to the host's profile picture.
• neighbourhood_cleansed: The specific neighborhood where the property is located.
• latitude and longitude : The geographical latitude and longitude coordinate of the property (used for map tool).
• property_type: The type of property being offered (e.g., apartment, house, etc.).
• room_type: The type of room being offered (e.g., entire place, private room, shared room).
• accommodates: The number of guests the property can accommodate.
• bathrooms: The number of bathrooms available to guests.
• beds: The number of beds available to guests.
• price: The nightly price for the property.
• numbers_of_review: The total number of reviews the listing has received.
• review_scores_rating: The overall rating score of the listing based on reviews.
• reviews: A list of individual reviews left by guests, each including the reviewer's name, date of review, and the review text.

The dataset was built from the data available on Inside Airbnb in the London section by using the listing.csv and reviews.csv files. After appropriate operations, these files were merged and transformed into approximately 39,000 JSON files.

Accommodations with null (or NaN) values and those without reviews were removed.

Authors

• Alberto Nuzzaci
• Luca Lodesani
• Riccardo Neri