BOTTARI: An augmented reality mobile application to deliver personalized and location-based recommendations by continuous analysis of social media streams
Introduction
Imagine that you are a tourist in Seoul. You would like to dine out. You prefer to avoid tourist traps and dine where the locals do. You have been told that Insadong district would be the perfect place; it offers a choice of more than a hundred restaurants in two square kilometres, and most of the district is reserved for pedestrians.
When you reach Insadong-gil (the main street of the district), you find yourself surrounded by hundreds of restaurant advertisements (see Fig. 1). You know that you can still open your guide book and choose one of the few restaurants listed there, but you definitely want a place where the locals go. You take out your mobile and check various apps that recommend restaurants based on users’ reviews. The number of user-rated restaurants is smaller than you expected: only ten restaurants are rated more than ten times. This is probably because you are in Seoul, one of the cities world-wide in which people tweet a lot.1 You wish that a service existed that continuously analysed the social media streams and that could show you how the locals have been rating Insadong’s restaurants over the last few months.
This is exactly what we designed BOTTARI for. BOTTARI is an augmented reality application for personalized and localized restaurant recommendations, experimentally deployed in the Insadong district of Seoul. At first glance, it may appear like other mobile apps that recommend restaurants, but BOTTARI is different: BOTTARI uses inductive and deductive stream reasoning [1] to continuously analyse social media streams (specifically Twitter) to understand how the social media users collectively perceive the points of interest (POIs) in a given area, e.g., Insadong’s restaurants.
In this paper, we describe the choices we made in designing BOTTARI and the lessons we learned by experimentally deploying it in Insadong. The paper is organized as follows. Section 2 introduces relevant background. Section 3 illustrates the BOTTARI mobile app from the user’s point of view, i.e., the main task being pursued. Section 4 explains how to understand the data used in experimentally deploying BOTTARI in Insadong district. Section 5 briefly illustrates the ontology at the core of BOTTARI that was used to integrate the available information. Section 6 presents the BOTTARI back-end. Sections 7 Evaluation, 8 Scalability report our evaluation results both in terms of quality of recommendations and scalability of the BOTTARI back-end. Finally, Section 9 concludes the paper by discussing the lessons we learnt and sketches our future works.
Section snippets
Background work
In this section, we briefly illustrate the context in which the idea of BOTTARI was conceived and the technological ingredients we used in implementing it.
The BOTTARI mobile app
As shown in Fig. 2, BOTTARI is an Android application (for smart phones and tablets) in augmented reality (AR) that directs the users’ attention to restaurants and dining places in the neighbourhood of their position.
In the Korean language, “bottari” is a cloth bundle that carries a person’s belongings while travelling. BOTTARI carries the collective perceptions of social media users about POIs in an area and uses them to recommend POIs. As shown in the upper-left corner of the screenshot in
Datasets used in BOTTARI
BOTTARI is built on two types of data: the static descriptions of the POIs and the social media streams.
Ontology used in BOTTARI
We designed BOTTARI following an ontology-based information access architecture [29]. The BOTTARI ontology is represented in Fig. 4. It extends the SIOC vocabulary [30], defining TwitterUser as a special case of UserAccont and the concept of Tweet as being equivalent to Post. It models the notion of a POI as NamedPlace extending SpatialThing from the W3C WGS-84 vocabulary.14 A NamedPlace is enriched with a categorization (e.g., the ambience describing the
Architecture and components
The BOTTARI architecture is illustrated in Fig. 5. It consists of three parts: (a) a client (described in Section 3) that interacts with the user and communicates to the back-end sending SPARQL queries, (b) a data initiated segment (PUSH) that continuously analyses the social media streams, and (c) a query initiated segment (PULL) that uses the LarKC platform to answer the SPARQL queries of the client by combining several forms of reasoning.
Evaluation
The quality and the efficacy of BOTTARI recommendations was comparatively evaluated using the data set described in Section 4.
Scalability
The Social Media Crawler (see Fig. 5) probes hundreds of thousands of tweets/day, but the RDF stream produced by the Opinion Miner contains an average of 150 RDF triples/day (corresponding to 75 tweets). The large majority of the crawled tweets are not related to Insadong’s restaurants. The flow rate of this RDF stream does not stress the PUSH segment of the BOTTARI back-end that runs on a laptop with CPU 2.8 GHz Intel Core i7 and 8 GB RAM DDR3, which corresponds to a share in a
Conclusions and future work
BOTTARI is a sophisticated application of semantic technologies that makes use of the rich and collective knowledge obtained by continuously analysing social media streams. We believe it was important to hide this complexity from the user using an intuitive and easy to use interface. The preliminary experiments we conducted show that BOTTARI can be more effective than guide books and Web 2.0 travel review sites.
Inspired by the literature on ontology-based information access, we designed the
Acknowledgment
This work was partially supported by the LarKC project (FP7-215535).
References (34)
- et al.
Deductive and inductive stream reasoning for semantic social media analytics
IEEE Intell. Syst.
(2010) - et al.
A large scale study of wireless search behavior: Google mobile search
- et al.
Deciphering trends in mobile search
IEEE Comput.
(2007) - et al.
Deciphering mobile search patterns: a study of yahoo! mobile search queries
- et al.
Voting with your feet: an investigative study of the relationship between place visit behavior and preference
- et al.
Improving local search ranking through external logs
- et al.
Location-aware click prediction in mobile local search
- et al.
Fancy a drink in canary wharf?: a user study on location-based mobile search
- et al.
Context-sensitive information retrieval using implicit feedback
- et al.
A large-scale evaluation and analysis of personalized search strategies
It’s a streaming world! reasoning upon rapidly changing information
IEEE Intell. Syst.
Data Stream Management: Processing High-Speed Data Streams
The power of events: an introduction to complex event processing in distributed enterprise systems
Querying RDF streams with -sparql
SIGMOD Rec.
Ep-sparql: a unified language for event processing and stream reasoning
Answer set programming for stream reasoning
Cited by (57)
Streaming linked data: A survey on life cycle compliance
2023, Journal of Web SemanticsAn authoring platform for developing smart apps which elevate cultural heritage experiences: A system dynamics approach in gamification
2024, Journal of Ambient Intelligence and Humanized ComputingEvaluating View Management for Situated Visualization in Web-based Handheld AR
2023, Computer Graphics ForumStreaming Linked Data: From Visionto Practice
2023, Streaming Linked Data: From Visionto PracticeAugmented Reality in Cultural Heritage: An Overview of the Last Decade of Applications
2022, Applied Sciences (Switzerland)