Unreal Estate Final Report
Clive Chen Emmanuel Kozman Gagandeep Nain
Table of Contents
Introduction 2
Context 2 Project Purpose 7
Project Design 8
Overview of Components 8 Diagrams 11 UML (use link for full image) 11 ER Diagram 12 Software Architecture Diagram 13 User Scenarios 14 All Users 14 Renter 15 Advertiser 16 Sequence Diagram 17
Review Contribution 18 Search Contribution 21 Booking Contribution 23 User Account Contribution 25 Conclusion 27 References 28
Appendix 30
Installation manual for windows 30 Languages and Technologies 30 Source code structure 30
Introduction
The Unreal Estate team was largely successful in producing an innovative and creative accommodation platform. The final design is one that leverages the current best practices for accommodation listing and booking as well as providing a new approach to managing and authenticating user reviews. This structure of this report details the culmination of information regarding the initial motivations, intermediate methods and final product of this project.
The motivations and context of the project will be described in the introduction and will present research conducted for the project proposal as well as extended research into user reviews. The exploration into reviews was forecast in our initial proposal scope, however, the degree of focus on this area was increased as the project progressed. After describing the background and history of accommodation platforms, the introduction will include the aim of the Unreal Estate project team which arose as a response to the current status quo.
The main body of the report details the outcome of this aim and the Unreal Estate’s final product. This will be achieved by first an overview of the system’s architecture, descriptions of the usage scenarios for all types of users and through various diagrammatic representations of the platform. This overview will be followed by a detailed description of the four main contributions of the website, the review component, the search component, the booking component and the user management component. These detailed contribution sections will provide deeper insight into the backend techniques that were used for each component as well as images of the frontend interface.
The report will finally conclude with an evaluation of the overall project. This will describe how the underlying fundamental booking system was completed well and how the review system could be further expanded. Furthermore, the challenges faced, as well as areas for improvement, will be outlined.
Economic Significance of This Project
The tourism industry represents around 10.4% of global GDP with a growth of 3.9% from 2018, second only to manufacturing. The World Travel & Tourism Council’s (WTTC) report also outlines the fact that the tourism industry accounts for $223bn of Australia’s GDP. The sheer economic scale and rapid growth of this sector indicates the importance of understanding and exploring the key structures that enable such an industry to operate. One particular key structure is the accommodation industry which allows for the free and convenient movement of people across regions. This report will analyse online accommodation booking in terms of what
makes an effective platform and propose its own online accommodation booking project which attempts to replicate and extend current examples.
Success of Online Travel Agencies
The success of Online Travel Agencies (OTA) is due to several factors including, distribution, availability, removal of barriers to entry and affordability. These factors are all enhanced forms of marketing or unique economic advantages afforded by the tool of the internet. Another key reason for success is the ability for everyone to give peer reviews.
The first and most obvious reason for the success of OTAs and one that translates to any online service is the power of increased distribution. Distribution refers to the specific method of delivery of products and services to customers. It is no surprise then that access to the internet into peoples homes and pockets has allowed a revolution in distribution. Consumers are now able to access the accommodation booking service more often and with greater convenience.
The removal of barriers to entry for both consumer and provider has been another revolution for OTAs. The most recent model for OTAs has been the peer-to-peer model, where individuals are able to become their own agent and be in contact with renters directly. This had led to a much wider adoption and acceptance of the online service. Coupled with the removal of barriers is the increased availability and quantity of accommodation in the market. Availability is also increased by increasing exposure from aggregating accommodation options in order to advertise more completely for a consumer.
Another consequence of increased availability is that it generates more affordable accommodation simply due to increased competition. In addition, the peer-to-peer model introduces unconventional, cheaper forms of accommodation such as a single bedroom in shared apartments, or famously, an airbed in a living room, as was the original idea behind AirBnB.
The final feature of being able to provide peer review has been important in allowing users to widely adopt OTAs. The personalisation and humanisation of the platform leads to customer retention and more trustworthiness.
These factors all need to be considered when analysing and creating a successful OTA platform as they are what distinguish it from traditional forms of accommodation services. In deciding the implementation of a specific online booking system it is important to understand the possible business models available.
History of Online Accommodation Bookings
Historically, the online accommodation booking industry has been around almost as long as the internet itself. The first notable online accommodation service was Travelweb.com, established
in 1994. It started off as simply an online hotel directory but functionality to make bookings added soon after. Many of the other major OTAs were founded within the next decade, with the AirBnb being a unique exception.
Online Tourist Agent Established
Travelweb.com 1994 Booking.com 1996 Expedia 1996 Webjet 1998 TripAdvisor, Inc. 2000 eDreams 2000 Wotif 2000 Opodo 2001 Trivago 2005 HomeAway (stayz.com.au) 2005 AirBnB 2008
This is clearly still a maturing industry which has seen different business models throughout its history always with area for improvement. There are four major models for OTAs, the advertising model, the agency model, the merchant model and the peer-to-peer model. Each has its own distinct set of advantages and disadvantages and also their own target demographics. The Unreal Estate team will consider the nature of each of these models in determining how to implement our own version.
OTA Business Models
The advertising model is one that assumes the lowest amount of risk. In this model, an OTA will not make any bookings but rather advertises rooms and redirects to other OTAs who handle the booking process using one of the other three models. The advertiser is then paid for the click-through to the site regardless of whether or not a booking is made. The most successful example of this model is Tripadvisor.com.
The agency model is perhaps the most conventional as it is the most similar to non-OTA methods. In this model, an OTA is awarded commissions on the sales that it makes on behalf of the hotels. The prices presented to the consumer are decided by the hotel and the OTA doesn’t
CS Help, Email: tutorcs@163.com
take on risk much risk by simply acting as an intermediary. The most prominent proponents of this model are Booking.com and Expedia.
The merchant model, which is arguably more common, is one where the OTA will purchase the right to rent a set of rooms from a hotel. This kind of wholesale model can be seen in other industries where the OTA will have a contract with the hotel to purchase the rooms at a discounted rate and then has the right to decide on what they will charge to customers. The OTA, in this case, assumes the risk if they are unable to rent out the rooms in exchange for flexibility with pricing. An example of an OTA that uses this model is Webjet.
The final model and perhaps the most disruptive to the industry is the peer-to-peer model. As mentioned earlier, this model allows individuals to act as their own agent for their own properties by connecting directly to renters. The OTA, in this case, acts as an intermediary for advertising, booking and conflict resolution. The original and most successful example is, of course, AirBnB. It will be this model that our project will aim to emulate.
Existing platforms
The current state of the online accommodation booking market capitalises on much of the reasons for success mentioned above. The distribution is inherent with any OTA, while the benefits of removing barriers for entry arise most directly from a peer-to-peer model. The affordability and availability will result from the market forces once a peer-to-peer platform is created.
In terms of the design-flow for existing OTAs, they all follow a similar pattern. The current platforms attempt to make booking as seamless as possible by requesting an input of minimal key search criteria, often being: date of departure, date of arrival, number of guests and location(s). The user is then presented with a list-view or map-view of properties within a small range of the specified location(s) with basic information about price, address and images.
These first few steps are vital for customer conversion, that is, getting the customer from the start of the website to completion of an actual booking. For the majority of websites, the customer does not need to login, they are simply looking for information. If a login was required, most customers would be discouraged from continuing any further thus reducing their website’s conversion rate. However, the lack of login means that there is a possibility for missed opportunities for gaining intelligence about customer behaviour when using the website. This might be data about how the customer
Once a customer has made a booking, there is minimal intervention from the website. The renter and host are able to exchange messages but overall there isn’t much further interaction. The host usually provides renters with a set of rules and the OTA only acts as conflict resolution. This minimal intervention approach is common among OTAs but again represents a possible source of missed opportunity to provide the renters and hosts with value or service.
The Effect of User Reviews
The introduction of user reviewing has had a major impact on the accommodation industry. The equalizing nature of a peer-to-peer system has meant that everyone user on these forums is able to express and evaluate their experiences to the rest of the user population. According to Gu et. al (2009) there is a significant r elationship between user reviews of hotels and their financial performance as a business. This is perhaps unsurprising as personal experience will inform you that you are heavily influenced by ratings and reviews for products that you view online. Interestingly, however, there is an asymmetry between positive and negative ratings (Avello et al. 2017). If a product has an overall positive rating, then the persuasiveness depends only on if there is a high total of reviews. However, if there is an overall negative rating, then there is no correlation between how effective it is on a user and how many negative reviews there are, that is, only a few are sufficient to dissuade.
The landscape of reviews is complicated further as they are often created as a strong emotional response to an experience. It has been suggested (Gupta et al. 2011) that negative reviews with emotionally charged language are often less compelling as users often attribute the review to an irrational disposition of the reviewer rather than the product. Conversely, positive reviews are taken to apply to the product rather than the emotional character of the reviewer. The informativeness of a review is, therefore, a complicated issue and one suggests that reviews are not taken simply at face value.
Project Purpose Aim
The Unreal Estate team set out to use what has been learnt and successfully applied by many of the leading accommodation platforms such as Trivago and AirBnB. The goal was to create a peer-to-peer accommodation rental platform with an increased focus on personalisation and user autonomy. Ultimately, the purpose of this project was to explore potential minor but important improvements to the online accommodation industry.
The scope of the work was initially to create a strong basic system with extended focus on personalised features. Therefore in terms of the fundamental features of an accommodation system our platform started with the property management component which includes key functions for listing, editing, and deleting properties. There is also the ability for both advertisers and renters to cancel bookings to allow for autonomy over their experiences and properties. Users have the ability to leave reviews and ratings on any properties stayed at which can be viewed by other users. In terms of personalisation, booking users have a route generated from their home to the property on booking to provide them with convenience.
The Unreal Estate project completed the basic system and progressed into extending the review capacity to address the issues mentioned above. The platform allows for advertisers to perform machine learning evaluations on the reviews of their properties in order to combat unfair and inconsistent reviews. There is an overall noise removal and smoothing effect of this machine learning which will be discussed in the following overview and in depth in the review contribution section.
Project Design Overview of Components
Advertising
Advertising module is designed for advertisers for property listing. After login, the owners will be able to add their properties for rentals. The module will allow them to add or update images, price, address, building type, number of rooms, number of beds, number of bathrooms and the number of guests allowed per room.
● Advertising page will not available unless the user is logged in
● After login, the owner can navigate the sidebar and be redirected to the advertising
● From advertising (owned property listing) page, the user can add new properties by
clicking “Add New Property” button at the bottom of the list.
● User is able to fill the form (including upload images) and click “Continue” to be
redirected to preview page or return to the previous page by clicking “return” (but the form will not be saved). It may take some time to be redirected to preview page if the image is huge.
● From the preview page, the user is able to add the new property by clicking the “confirm” button
● From the advertising page, the user is able to update property by clicking “Update” beside each property in the list. User will be redirected to the form.
● From the form, property information is preloaded, and the user is able to update the information and submit the form by clicking “Update”. Or the user can return to the previous page by clicking “return” (but the new changes will not be saved).
● From the advertising (owned property listing), user can delete the property by clicking “DELETE” button beside each property in the list.
The user will have the ability to review a property after their period of accommodation has ended. This will be a rating between 1 – 5 in terms of stars in increments of 1. Along with this, the user will get to share their feedback and review in a text box.
This review will be attached to the user, the owner of the accommodation, and the accommodation itself. This means that if another user is viewing the property, they will see all reviews for that property and the average rating of that property.
The user will also be able to visit a page with all their bookings in their account, and for each, if the period of accommodation is over, then the user will be able to submit a review, however, if they already have, then they will only be able to view the review. When leaving a review, the user will be anonymous in their review such that the owner and other users will not be able to see who left the review.
The property manager can go into the ‘My properties’ page, and then for each property, they can view the reviews on that property. They can then use our inbuilt machine learning functionality in order to detect any misleading reviews where the rating appears to be different from the overall tone of the review text. Using the machine learning network, the property manager can choose to replace any mismatched reviews if the difference in rating between the user and the AI is greater than or equal to 2 stars. However, this needs to be done for all the reviews on the property since we don’t want the property manager to be able to pick and choose higher ratings in order to boost the property’s overall average rating. The details of this module are explored further in the following section in contribution 1.
The Accommodation Search Module is for users to make detailed searches on properties. The visitors will be able to choose accommodation preferences and receive a list of suggestions. Features:
● A search bar for the visitor to input locations(street, unit number, suburb, state, postcode), with autocomplete feature that will try to match the input to an address using the Google Maps API.
● User is able to choose the start date and end date of accommodation
● User is able to enter the number of guests that will be staying at the property.
● The result will be a list of properties within 1km of the search destination, the correct
number of guests, and availability within the date period. With further information that the
user can navigate with.
● The search results are paginated such that only a subset of the results are displayed at
first, then the user can choose to view more by scrolling to the bottom and clicking to
load more results.
● Search results will show all essential qualities of a property, including the title, an image,
the average review score, the suburb, number of bathrooms, bedrooms, and number of guests allowed, price, and the building type.
The information presented to the user throughout the searching process should be simple and relevant in order to assist them to make the correct choice about the properties they would like to view. This module will be described in more detail in the Search Contribution, in the following section of this report.
Naturally, a fundamental aspect of all accommodation platforms is the ability to make bookings. The Unreal Estate platform allows authenticated users to make bookings with any of the listed properties. Booking a property is done via the search function. Once a property is selected from the search, the user will be presented a detailed summary of the property. If a user is not logged in, there will be a button at the bottom which redirects the user to login. If the user is logged in this button will allow the user to continue to a booking form.
● Booking form allows the user to modify the originally searched check-in and check-out
● Booking form allows the user to modify originally searched number of guests (to a
maximum of the properties allowed guests)
● Booking form presents the total price
● Bookings are able to be cancelled by both owners and renters after a booking has been
This module is detailed in the following section of the report in the Booking Contribution.
User Account Management
The user account management module describes the component of the website which handles all user authentication. This module is key for every other module apart from basic searching (which still depends on users being able to advertise). All users who want to interact with other users need to be associated with an account. Each account includes their personal details as well as their trips, their properties and the bookings associated with each property.
● Users authenticate with a username and password
● The username is equivalent to their email address
● Users have access to past, present and future bookings they make
● Users have access to past, present and future bookings made with their properties
● Users have access to list their own properties
● Users have access to review properties they have booked
This module is explained with images in the User Account Contribution section.
UML (use link for full image) https://drive.google.com/file/d/1uFgrQEfz3xBNzXeEKdLPzRrccEEi1Lbp/view?usp=sharing
ER Diagram
Software Architecture Diagram
User Scenarios
The main users are renters and advertisers. Once the homepage is loaded, a request is sent to Google Maps API to fetch the current location of the user based on their IP address. This information is used to present users with default properties on the homepage before a search has been made.
Before logging in, both renters and advertisers need to make sign-up a nd sign-in requests. The sign up r equest is made with function EnterSignUpDetails to create an account. Then, when the user tries to log in, a request is handled by CheckUserDetails f unction which queries the AWS Cloud Database, and it will respond as either failure or success. Upon a successful response, the user is logged in.
After logged in, a renter can fill the search fields on the homepage and make a search. During this process, requests are sent to google maps API for address autocomplete.
After the renter clicks search, a request is sent to the database, and the PrimarySearch() function will filter out the irrelevant results and return a list of properties, which is then displayed on the search results page.
When the user confirms the booking, a request is sent and the function BookingFunction() in the booking model will create a new booking. In the meantime, a request is sent to the Google Maps API to display the route from the current location to the property location on a map.
When a user clicks Bookings o n the sidebar, a request will be handled by the function bookingDetailsBID() to fetch the booking object by booking ID (BID).
Advertiser
After logging in, the advertiser can request for the list of owned properties by clicking the “My Properties” button on the sidebar. This request will be handled by function list_property() in advertising model, which returns a list of properties or an empty list. Request for deleting, updating and adding new property are handled by PropertyFunction() in advertising model.
程序代写 CS代考 加微信: cstutorcs
Sequence Diagram
Review Contribution
There are two parts to this component
1) Adding a review to a booking
2) Classifying a rating based on the text in the review
Adding a review is done by first having rented some accommodation, and then after the rental is over, the user will be able to review their stay from the Bookings page (accessible from the side menu). This review is made up of a rating from 1 -5 stars (increments of 1) and a text field, where the user is able to share more about their experience as shown in the image.
This input will be added to the database along with the booking_id, property_id, and user_id it is associated with as a review. These reviews can then be seen on the Property page, as well as the overall ‘average’ of the reviews.
If a user wants to view their review after leaving it, they can do so by clicking ‘View Review’ from Bookings (this will change from Write Review after a review has been left). The Property Manager can also view all reviews for any of their properties in order to keep up to date with the user’s experience with their property.
The other part is that the property owner can also utilise our inbuilt machine learning algorithm to detect reviews that have ratings that are contradictory to the overall sentiment left in the text review. After this detection the property manager can choose to overwrite the User rating with the Neural Network’s rating however, this will only be done on Reviews that have a difference greater than or equal to 2 stars between the User and AI rating. Also, this can not be done selectively i.e per review and must be done to all reviews on the property as to prevent the property manager from just using this feature selectively when it increases their properties average rating.
This is done by retrieving the property_id of the selected property, then the backend will retrieve all reviews on that property, and feed each review into the neural network classifier. The classifier will make predictions on each review based on the context of the text, and return a
Programming Help, Add QQ: 749389476
rating between 1-5 stars. We can then return this to the frontend in order to display the neural network’s ratings, where the property manager can then make the decision whether or not to use the user or network’s ratings.
Here is an example below of two very misleading reviews, where in the first case, the User gives a 5-star rating, however, proceeds to write very negative feedback based on their experience. The trained neural network is able to detect that this should indeed be classified as a 1-star rating. In the second case, the User gives a 1-star rating, however, proceeds to write very positive feedback based on their experience. Again, our neural network is able to detect that this review should have been classified as a 5-star rating.
Now the property manager is able to use the button to use the AI’s ratings instead, and while this may not increase the property’s average rating, it will lead to more consistent reviews, and filtering out fake reviews.
Another change that was made to reviews in order to increase the reliability of an average rating, was to use a ‘trimmed mean’ in which we filter/trim out the top and bottom 10% of ratings such that any outliers would not be accounted for in the average. This turned out to be a good method as not many properties have more than 10 reviews, and hence most properties have a real average, however, the properties that do have more than 10 reviews have a much more consistent average if there is a case of outliers.
Both of these features are implemented in order to help both renters and owners to have more consistent reviews at a glance, which in turn may persuade/dissuade users accordingly, and help owners get the feedback they need in order to improve their properties/service.
The Neural Network model was made using TensorFlow. The model is a bidirectional recurrent neural network, where each directional layer is multilayered, and made of two LSTM layers. Both directional layers are the same in that they have a layer of 50 LSTM cells, feeding into 100 LSTM cells. Some training choices include training in batches of 100 reviews, using dropout of 0.6 for a more flexible network, a small learning rate of 0.005 such that learning doesn’t blow up.
We trained the model on an unofficial dataset of TripAdvisor reviews, which was scraped from their website, and can be found here (https://www.cs.cmu.edu/~jiweil/html/hotel-review.html). This did involve some challenges such as reviews in different languages, however, this was avoided by filtering out reviews in languages other than English. We ended up with 45000 total reviews per star from 1-5 and then split these 45000 into an 80/20 split for training/testing.
We also used the word to vector embeddings by GloVe, this will be automatically downloaded the first time you use the classifier on the website (as the file was too big to push to Github). For more information, you can visit (https://nlp.stanford.edu/projects/glove/).
Our neural network ended up with an accuracy of ~63% as shown in the figure below, which is better than some other models that have been used to do this task, such as the following: http://www.developintelligence.com/blog/2017/03/predicting-yelp-star-ratings-review-text-python /
Search Contribution
The Accommodation Search module was built so that the user can efficiently and intuitively search for their desired accommodation without having to login/signup such that they can simply check what properties are available without the hassle.
We have added some additional features on the frontend to make searching as effortless as possible, such as the automatic translation of the user’s input to an address as shown in the image below. Once the user begins to type a location, we have incorporated the Google Places API in order to translate this text, to some popular places that match the input so far. As can be seen in the image, the frontend will display 5 potential places that are closest to your input as determined using the API, after which, the user may simply select one, and the address will be auto-filled. This makes it much easier to input an address for the user, especially if it is a popular location.
When selecting the check-in and check-out dates on the search page, they have been restricted such that the user is not able to input invalid data, so that it is as easy as possible for users to enter the correct dates. Some of these restrictions include, not being able to choose a check-in date previous to the current date, and not being able to choose a check-out date before the check-in date. An example of t