MISTRuST: accoMmodatIon Short Term Rental Scanning Tool

an Ruiz-Rube

1 a

, Inmaculada Arnedillo-S

anchez

2 b

and Antonio Balderas

1 c

School of Engineering, University of C

adiz, Spain

School of Computer Science and Statistics, Trinity College Dublin, Ireland

Keywords:

Accommodation Rental Platforms, Machine Learning, Web Scraping, Search Engines.

Abstract:

The global irruption of ‘shared-accommodation platforms’ has ignited debate regarding the implications of

the unprecedented growth of the short-term rental market. While some argue that they have generated new

business, work and wealth others, highlight their wider societal, economic and legal effects. Short-term rental

removes long-term housing from the market, forces rent prices up, saturates areas with tourism, generates

safety and liability concerns and by and large, it is awash with likely illegal listings. This paper presents

MISTRuST (accoMmodatIon Short Term Rental Scanning Tool) a computational intelligence based system

aimed at uncovering whether a property is being listed in short-term rental (STR) platforms. It enables users

to monitor property by scheduling automatic searches and checking listings returned by the system against the

target property. The asynchronous pipeline architecture involves three stages: data ingestion, data enrichment

and data matching. Preliminary test results with a set of listings are encouraging. However, further evaluation

is needed to improve the accuracy of the system on a larger scale. It is hoped MISTRuST will help stakeholders

such as property owners, state and agencies and others tackle the growing concern over unlawful STRs and

contribute towards sustainable solutions.

1 INTRODUCTION

According to Schort (Schor, 2014) the launch of

the sharing economy was marked by positive mes-

sages on how technological and economic innova-

tion would provide ﬁnancial beneﬁts to ordinary peo-

ple. In this vein, the shared-accommodation platform

Airbnb, claims it helps individuals generate extra in-

come from their underutilised accommodation to pay

their rent or undertake a home renovation project.

Founded in 2008, Airbnb was valued at 31 billion

U.S. dollars in May 2017 and has over 3 million list-

ings in 190 countries and 65,000 cities. These ﬁg-

ures endorse the role of the company as an enabler of

economic growth (Rae, 2018) through its direct op-

eration and by indirectly generating tourism-related

jobs (Nieuwland and van Melik, 2018). However, as

demand for short-term rental grows, long-term rental

is removed from market to meet that demand and

rent prices of the diminishing stock are rising push-

ing people out of the rental market (Wachsmuth et al.,

2018). Beyond the ﬁnancial implications of this phe-

https://orcid.org/0000-0002-9012-700X

https://orcid.org/0000-0002-8749-9611

https://orcid.org/0000-0003-0026-7410

nomenon, the fabric of neighbourhoods is experienc-

ing unprecedented changes. Long-term neighbours

are being replaced by short-term visitors (Wachsmuth

et al., 2018), bakeries, greengrocers and butchers by

souvenir shops and residents are sieged by battal-

ions of foreigners to the conquest of their neighbour-

hoods armed with their trolley cases in hand. While

those engaging in short-term rental masquerade be-

hind platforms and beneﬁt from the overwhelming

regulatory vacuum in the sector, local residents en-

dure overcrowding, anti-social behaviour, safety con-

cerns (Gurran and Phibbs, 2017) and injurious lack

of enforcement of conveyance, planning permission,

local and governmental regulation and common law.

Most jurisdictions outlaw the use of residential ac-

commodation for commercial short-term rental (Katz,

2015). To this end, the most comprehensive study to

date on the issue, reports that 66% of revenue ($435

million) and 45% of all New York Airbnb reserva-

tions last year were illegal because it breached New

York State law (Wachsmuth et al., 2018). Never-

theless, the proliferation of STR platforms such as

Airbnb.com, Booking.com, Homeaway.com, Home-

stay.com, Couchsurﬁng.com, to mention just a few,

makes enforcement an insurmountable task. For

170

Ruiz-Rube, I., Arnedillo-Sánchez, I. and Balderas, A.

MISTRuST: accoMmodatIon Short Term Rental Scanning Tool.

DOI: 10.5220/0010104801700177

In Proceedings of the 16th International Conference on Web Information Systems and Technologies (WEBIST 2020), pages 170-177

ISBN: 978-989-758-478-7

starters, unless privy of listing data, often conﬁden-

tial to the STR platforms and person offering the

accommodation, it is difﬁcult to locate speciﬁc list-

ings among millions and assert they are legal or ille-

gal. Being able to monitor listings on STR platforms

should be of interest to authorities and local residents

seeking enforcement of regulation but also, to land-

lords to ensure their property is not being used for

STR without their knowledge. The STR subletting

phenomenon is a real threat to landlords that can have

serious consequences as, for instance, being ﬁned by

local authorities and been unable to evict tenants en-

gaging in STR subletting. This paper presents MIS-

TRuST, a computational intelligence based system,

aimed at uncovering whether a property is being listed

in short-term rental (STR) platforms. It enables users

to monitor property by scheduling automatic searches

and checking listings returned by the system against

the target property. A methodology based on the de-

sign and creation strategy (Oates, 2005) and an agile

life-cycle were adopted for its development. The rest

of the paper is structured as follows. The background

and related work are presented in Section 2. The re-

quirements and features are described in Section 3.

Section 4 elaborates on the asynchronous pipeline ar-

chitecture involving three stages: data ingestion, data

enrichment and data matching. Finally, section 5 pro-

vides discussion and future work.

2 ADVERTISEMENT LISTING

DATA RESEARCH

As outlined previously, there are regulatory, ﬁnan-

cial, and societal implications of prevailing use of

STR platforms. Exploring potential public policy re-

sponses may have in rental prices, Filippas (Filippas

and Horton, 2017) entertains four scenarios: the deci-

sion to rent resides with the individual host, the build-

ing owner, the city or social planner. Although policy

choices seem to have no detectable effect on rental

prices (Filippas and Horton, 2017), ‘algorithm regu-

lation’ would enable policymakers to draft regulations

relying on data analysis to be responsive to real-time

(Quattrone et al., 2016). Thus, gathering information

from the online platforms and matching listing infor-

mation with census and hotel data, it would be pos-

sible to determine the socioeconomic conditions of

the areas that would beneﬁt from the hospitality plat-

forms. Personal characteristics may inﬂuence STR.

Xiao (Ma et al., 2017) examine how hosts describe

themselves on their Airbnb proﬁle pages and exam-

ine their perceived trustworthiness. Along the line of

trust and veriﬁcation, Zhang (Zhang et al., 2016) uses

machine learning techniques to analyse the impact of

having listings’ photos veriﬁed. Results illustrate that

listings with veriﬁed photos are 9% more frequently

booked.

In addition to the previous, attempts have been

made to quantitatively characterize collaborative con-

sumption behaviors in Airbnb (Lee et al., 2015) and

analyze the economic value of trust artefacts (Teub-

ner et al., 2016). Social features, such as responsive-

ness of host, number of reviews, membership senior-

ity and ‘Super-host’ status, are signiﬁcantly associ-

ated with room sales, providing economic value (Lee

et al., 2015; Teubner et al., 2016).

There is a lack of peer-reviewed literature re-

garding property matching on STR platforms, except

for a couple of commercial service providers, suble-

talert.com and subletspy.com, whose marketing sum-

maries could be similar to our proposal. However,

they do not disclose any technical details of their im-

plementation and they do not seem to provide any

inherent scientiﬁc contribution. However, there is

work on the usage of recommender systems to help

prospective tenants ﬁnd a ﬁtting property (Yuan et al.,

2013).

3 MISTRuST

(ACCOMMODATION SHORT

TERM RENTAL SCANNING

TOOL)

Since the publication of aggregated datasets of list-

ings from Airbnb in the portal insideairbnb.com (Cox,

2017), the number of studies analysing the impli-

cations of STR platforms for hotels, neighbourhood

cohesion and rental market have increased. The

increased interest in analysing listing data and un-

derstanding the multidimensional implications of the

STR phenomenon does not seem to be matched by an

equal interest in the design and development of tools

that help stakeholders automatically scan and monitor

properties listed in such platforms. Thus, we envis-

aged the design and development of an accommoda-

tion STR scanning tool.

3.1 Requirements

MISTRuST functional requirements involve (i) regis-

tration of the data of the property to be monitored; (ii)

issuing queries to check whether the property is listed

on any STR platform; (iii) setting and receiving alerts

once a potential match to the target property has been

found listed in STR platforms.

MISTRuST: accoMmodatIon Short Term Rental Scanning Tool

171

With regards to the non-functional requirements,

the following quality attributes were considered:

• Security. An authentication and authorization

mechanism was included to protect the conﬁden-

tial data of the users.

• Interoperability. The system connects with on-

line providers (alltherooms.com) of accommoda-

tion advertisements.

• Performance. A data cache module and a multi-

thread streamed data processing scheme was in-

cluded for proper performance.

• Reliability. An exception handling system and

data cleansing and enriching tasks are applied to

improve error tolerance.

• Portability. The system was developed with stan-

dards languages and formats, for compatibility

with a regular web browser and even via an API

provided for further processing.

• Usability. Common heuristics in user interface

design have been applied to provide an adequate

user experience.

• Maintainability: all the components of the system

were developed by using well-known design pat-

terns and principles in Software Engineering to

ensure its sustainable evolution.

3.2 System Walk-through

Once the user logs on MISTRuST, the system al-

lows access to three different modules, namely prop-

erty management, search management and property

ﬁnder.

From the Property Management view, the user can

edit the list of properties. For each property, the user

can register general information, upload pictures and

annotate the property entry with the amenities or fea-

tures it provides (see Figure 1).

Once the user has registered the properties, the

next step is to issue searches on the Internet. There

are two ways for doing that: (i) by selecting a speciﬁc

time interval (check-in and check-out dates) or (ii) by

scheduling the searches to be automatically launched

on a regular basis. In Figure 2, a summary of each

search issued for a given property is presented.

The results of the searches are incremental. Thus,

regardless of the number of searches issued for each

property, the listings collected are considered a single

set. Therefore, in order to check whether a property

is being listed, the user only has to select the property

from the selector in the property ﬁnder window. Af-

terwards, the system will display the list of listings

ranked by the computed score. The best-informed

guess of the system is ranked ﬁrst so that the user will

be able to check the details of the listing against the

features, pictures and amenities of the target property

(see Figure 3).

4 SYSTEM ARCHITECTURE

The architecture of the system which was carefully

designed to fulﬁll its requirements follows a three-

layered architecture along with an asynchronous

pipeline one (see Figure 4). The whole system was

developed with Java and Spring Framework.

The presentation layer consists of the different

views of the system and was developed with Vaadin, a

user interface framework for creating HTML5 single-

page applications. The service layer includes several

business components to manage the system entities

and to implement the core algorithm of the system.

The data layer was built on top of two data reposi-

tories, namely MySQL, for persisting the relational

domain model, and Apache SOLR, for indexing ad-

vertisement data in form of ﬂatten documents. The

core of the system follows an asynchronous pipeline

architecture. Three data processing stages are carried

out before providing results for a given query: (a) data

ingestion, (b) data enrichment and (c) data matching.

However, not all three stages have to be executed at

the same time.

4.1 Data Ingestion

The ﬁrst data processing step involves obtaining data

from Alltherooms.com. Since this web platform ag-

gregates data from several popular STR platforms,

such as Airbnb.com, Homestay.com, and Home-

away.com, the extraction process is limited to a sin-

gle web platform. However currently, this web plat-

form does not provide any structured API, so in order

to ﬁnd the listings the system has to directly scrape

listing data from its user interface. Thus, the system

issues an HTTP request to search for accommodation

near the location of the user’s property on the search

page Alltherooms.com.

The exact location of the listing is unknown be-

forehand in order to guarantee the personal rights of

the users. The webpage only provides us with the dis-

tance from the search point to the location of the list-

ing for those listing further than 1 mile away (see Fig-

ure 5). Thus, to estimate the real distance (d) from a

given listing (P) to the base position (O), two parallel

searches are launched in equally distant locations to-

wards West (A) and East (B). By applying the law of

cosines to the formed triangles, a system of equations

WEBIST 2020 - 16th International Conference on Web Information Systems and Technologies

172

Figure 1: Editing form of properties.

Figure 2: List of searches launched for a given property.

MISTRuST: accoMmodatIon Short Term Rental Scanning Tool

173

Figure 3: MISTRuST window showing the data of the user’s property (left) and a given ad (right).

Figure 4: Three-layered partially-pipelined architecture.

(1) and (2) is generated, whose solution (3) represents

the expected distance.

AOP : d

= d

+ (2R)

− d

(2R)cosα (1)

APB : d

= d

+ (4R)

− d

(4R)cosα (2)

d =

+ d

− 4R

(3)

In addition to the distance from the search point

to the listing location, the scraper fetches (for every

listing) the general description of the listing, its gen-

eral features (number of rooms, bedrooms, etc.), its

amenities and its pictures. Besides, due to the fact

that the web provider may return a considerable num-

ber of listings in a single query, the scrapper will it-

erate over each result page until a certain threshold

distance. The listing might not be available the whole

year but rather during different time intervals, for in-

WEBIST 2020 - 16th International Conference on Web Information Systems and Technologies

174

Figure 5: Estimation of the distance between the actual location of the property and that of a given advertisement.

stance, weekends or holidays. So, the scraping pro-

cess can be both launched on user demand for spe-

ciﬁc dates, or scheduled in advance for speciﬁc time

periods.

4.2 Data Enriching

Besides the approximated distance between the list-

ing and the property locations, their set of features or

amenities are essential to ﬁnd the best listing match.

Some amenities and features, such as elevators and

air-conditioning systems, are explicitly stated by the

user. However, there are other features which are

not deﬁned in a structured way but usually collected

through textual descriptions. Moreover, the pictures

uploaded by the users can also provide useful infor-

mation about other features of the properties. In order

to enrich the information collected from each listing,

the system makes use of the cloud machine learning

APIs provided by Google for processing natural lan-

guage

and pictures

. By analyzing the textual de-

scriptions and the images of the advertisements, the

system can obtain a considerable number of tags de-

scribing them. However, there is a huge variability on

the kind of entities which are automatically discov-

ered from the listings, such as house facilities, events,

street names, user’s rules, and so forth. So, a further

classiﬁcation process is required. The use of semantic

web technologies (Vandenbussche et al., 2017), such

as linked open vocabularies, can alleviate this task.

This system uses the Google Knowledge Graph

convert the tags into features and amenities according

https://cloud.google.com/natural-language/

https://cloud.google.com/vision/

https://developers.google.com/knowledge-graph/

to the Schema.org

vocabulary.

4.3 Data Matching

Once the listing data is downloaded and enriched by

means of the machine learning services, the ﬁnal stage

is the data matching. The open-source search plat-

form Apache SOLR is used for this. Unlike standard

databases, this software enables us to send and index

documents and to query them by using complex crite-

ria.

Firstly, initial ﬁlter criteria, such as the number

of bedrooms and bathrooms and the accommodation

type (ﬂat, house, cottage, etc.), are applied to re-

strict the superset of advertisements. Secondly, the

results are ranked by a computed score combining (i)

the computed approximated distance from the posi-

tion of the listing to the location of the real property

and (ii) the matching between the amenities/features

of the listing and those of the property. In order to ﬁnd

matches between the features and amenities of listings

with those of the properties, some aspects have to be

considered. On the one hand, not all the entities have

the same relevance. For example, a user’s rule, such

as ‘dogs allowed’ or ‘towels provided’, does not pro-

vide any useful insight, but amenities such as elevator

or heating do. For that reason, all the amenities can in-

clude a signiﬁcance factor between 0 and 1, indicating

its usefulness for the matching process. On the other

hand, because of the own nature of the machine learn-

ing models, it is necessary to consider a certain conﬁ-

dence value to their results. For example, in the case

of natural language recognition, the Machine Learn-

ing (ML) model provides us with information about

https://schema.org/

MISTRuST: accoMmodatIon Short Term Rental Scanning Tool

175

the relevance of each recognized entity to the full-text

description of the listing. Furthermore, the ML model

for image recognition returns a number representing

the accuracy of the entity detection. In the case of the

features and amenities explicitly deﬁned by the users

without the support of any ML model, it is considered

a conﬁdence factor of 1.0.

Both the relevance factor and the conﬁdence value

are hence taken into account to calculate the ﬁnal

score. However, before computing the score and rank-

ing the results, a query expansion process is applied

to deal with synonymous tags, parent/child terms and

translations in different languages. This process is

carried out by means of tables with recursive relation-

ships and the WordNet lexical database

(this step

is still under development). Finally, the results are

ranked by the score.

5 DISCUSSION AND

CONCLUSION

Short-term rental removes long-term housing from

the market, forces rent prices up, saturates areas with

tourism, generates safety and liability concerns and

by and large, it is awash with likely illegal listings. In

this paper, we have presented MISTRuST (accoMmo-

datIon Short Term Rental Scanning Tool) a machine

learning based system aimed at uncovering whether

a given property is being listed in short-term rental

(STR) platforms. It enables users to monitor property

by scheduling automatic searches and checking list-

ings returned by the system against the target prop-

erty. During the development of MISTRuST valid-

ity considerations have been contemplated. Firstly,

the manual allocation of the relevant values of the

amenities and features is a time-consuming task. The

required time can be reduced by assigning default

values to the categories they belong. Secondly, we

are subjected to the terms and conditions of the STR

platforms For example, Airbnb bans the use of au-

tomated data scrapping tools but Alltherooms.com

doesn’t. Nonetheless, this may change in the fu-

ture. In addition, Alltheroms.com doesn’t currently

provide an API so we are using web scraping tech-

niques. Any future changes in the user interface de-

sign of Alltheroms.com may impact our tool. Lastly,

a large-scale evaluation, with many properties and in

different locations would be necessary in order to ob-

tain accurate results. However, no software will be

able to ensure a 100% the accuracy of its guesses due

to the dependence on completeness of the property

https://wordnet.princeton.edu/

data provided by the user and the uncertainty of the

ML algorithms themselves. Our next step is to apply

a machine learning model to improve the ranking al-

gorithm for the top N retrieved listings, to avoid man-

ually adjust the relevance’s value of each feature or

amenity. To accomplish that, a substantial amount of

training data is required. So we will introduce a user

feedback system to register the success or the failure

of the algorithm executions and so to train the ML

model. Finally, we plan to release the software as

open-source.

ACKNOWLEDGEMENTS

The work has been supported by the University of

Cadiz with grant ref. EST2018-167 and funds of its

Department of Computer Engineering.

REFERENCES

Cox, M. (2017). Inside Airbnb: adding data to the debate.

Accessed: 2019-09-11.

Filippas, A. and Horton, J. J. (2017). The tragedy of your

upstairs neighbors: When is the home-sharing exter-

nality internalized? Available at SSRN 2443343.

Gurran, N. and Phibbs, P. (2017). When tourists move in:

How should urban planners respond to Airbnb? Jour-

nal of the American Planning Association, 83(1):80–

92.

Katz, V. (2015). Regulating the sharing economy. Berkeley

Technology Law Journal, 30(4):1067–1126.

Lee, D., Hyun, W., Ryu, J., Lee, W. J., Rhee, W., and Suh,

B. (2015). An analysis of social features associated

with room sales of Airbnb. In Proceedings of the 18th

ACM Conference Companion on Computer Supported

Cooperative Work & Social Computing, pages 219–

222. ACM.

Ma, X., Hancock, J. T., Lim Mingjie, K., and Naaman, M.

(2017). Self-disclosure and perceived trustworthiness

of Airbnb host proﬁles. In Proceedings of the 2017

ACM Conference on Computer Supported Coopera-

tive Work and Social Computing, pages 2397–2409.

ACM.

Nieuwland, S. and van Melik, R. (2018). Regulating

Airbnb: how cities deal with perceived negative ex-

ternalities of short-term rentals. Current Issues in

Tourism, pages 1–15.

Oates, B. J. (2005). Researching information systems and

computing. Sage.

Quattrone, G., Proserpio, D., Quercia, D., Capra, L., and

Musolesi, M. (2016). Who beneﬁts from the ”sharing”

economy of Airbnb? In Proceedings of the 25th Inter-

national Conference on World Wide Web , WWW ’16,

pages 1385–1394, Republic and Canton of Geneva,

WEBIST 2020 - 16th International Conference on Web Information Systems and Technologies

176

Switzerland. International World Wide Web Confer-

ences Steering Committee.

Rae, A. (2018). From neighbourhood to a globalhood?

three propositions on the rapid rise of short-term

rentals. Area, 0(0).

Schor, J. (2014). Debating the sharing economy. a great

transition initiative essay. Online: Great Transition

Initiative.

Teubner, T., Saade, N., Kawlitschek, F., and Weinhardt, C.

(2016). It’s only pixels, badges, and stars: On the

economic value of reputation on Airbnb. In Australian

Conference on Information Systems.

Vandenbussche, P.-Y., Atemezing, G. A., Poveda-Villal

on,

M., and Vatant, B. (2017). Linked open vocabularies

(LOV): a gateway to reusable semantic vocabularies

on the web. Semantic Web, 8(3):437–452.

Wachsmuth, D., Chaney, D., Kerrigan, D., Shillolo, A., and

Basalaev-Binder, R. (2018). The high cost of short-

term rentals in New York City. School of Urban plan-

ning, McGill University: Montreal, QC, Canada.

Yuan, X., Lee, J.-H., Kim, S.-J., and Kim, Y.-H. (2013).

Toward a user-oriented recommendation system for

real estate websites. Information Systems, 38(2):231

– 243.

Zhang, S., Lee, D., Singh, P. V., and Srinivasan, K. (2016).

How much is an image worth? an empirical analy-

sis of property’s image aesthetic quality on demand at

Airbnb.

MISTRuST: accoMmodatIon Short Term Rental Scanning Tool

177