Where Should I Go? A Deep Learning Approach To Per
Where Should I Go? A Deep Learning Approach To Per
Exam (elaborations)
Where Should I Go? A Deep Learning Approach To Personalize Type-based Facet Ranking for POI Suggestion
0 view 0 purchase
Course
Where Should I Go? A Deep Learning Approach To Per
Institution
Where Should I Go? A Deep Learning Approach To Per
1 Introduction
The problem of Point-of-Interest (POI) suggestion has gained lot of research
attention recently due to the spread of Location Based Social Networks (LBSN).
The problem is concerned with recommending interesting places for the users to
visit. POI suggestion algorithms personalize ...
Where Should I Go? A Deep Learning Approach
To Personalize Type-based Facet Ranking for
POI Suggestion
Esraa Ali1 [0000−0003−1600−3161] , Annalina Caputo2 [0000−0002−7144−8545] ,
Séamus Lawless1[0000−0001−6302−258X] , and Owen Conlan1[0000−0002−9054−9747]
1
ADAPT Centre, School of Computer Science and Statistics, Trinity College Dublin
esraa.ali,seamus.lawless,owen.conlan@adaptcentre.ie
2
ADAPT Centre, School of Computing, Dublin City University
annalina.caputo@adaptcentre.ie
Abstract. In a faceted search system, type-based facets (t-facets) rep-
resent the categories of the resources being searched. Ranking algorithms
are needed to select and promote the most relevant t-facets. However, as
these are extracted from large multi-level taxonomies, they are impos-
sible to show entirely to the user. Facet ranking is usually employed to
filter out irrelevant facets for the users. Existing facet ranking methods
neglect both the hierarchical structure of t-facets and the user historical
preferences. This research introduces a personalized t-facet ranking that
addresses both issues. During a first step, a Deep Neural Network (DNN)
model is trained to assign a relevance score to each t-facet based on three
groups of relevance features. The score reflects the t-facet relevance to
the user, the input query, and its general importance in the dataset. Sub-
sequently, these scores are aggregated and the t-facets are re-organised
into a smaller sub-tree to be presented to the user. Our approach aims at
minimizing the effort required by the user to reach their intended search
target. This is measured in terms of number of clicks the user has to
perform on the t-facet tree to reach a relevant resource. The approach is
applied to a Point-Of-Interest suggestion task. We solve the problem by
ranking the categories of the venues as t-facets. The evaluation compares
our DNN-based approach with other existing baselines and investigates
the individual contribution of each group of features. Our experiment
has demonstrated that the proposed personalized deep learning model
leads to better t-facet rankings and minimized user effort.
Keywords: Facet Ranking · Deep Neural Networks · Personalization.
1 Introduction
The problem of Point-of-Interest (POI) suggestion has gained lot of research
attention recently due to the spread of Location Based Social Networks (LBSN).
The problem is concerned with recommending interesting places for the users to
visit. POI suggestion algorithms personalize the recommendations according to
the current user’s context as well as the recorded user’s history.
, 2 E. Ali et al.
In this research, we solve the POI suggestion problem in the context of
Faceted Search Systems (FSS), where the categories of the POIs are used as
type-based facets (t-facets) that help the user to navigate the information space.
Literature showed that the categories of POIs play a key role in solving this
problem [2, 3]. FSS provide users with a set of t-facets to help them in filtering
and narrowing down search results and locating the intended POI quickly. When
POIs’ categories are derived from large, multi-level taxonomies, the FSS need to
implement methods to identify and prioritise the most relevant t-facet to show
to the users to avoid overwhelming them.
In this work, we focus on analysing the role of personalization in t-facet
ranking in isolation from other FSS aspects. We aim at answering the following
research question: To what extent a Deep Neural Network (DNN) model can
learn to rank t-facets in order to minimize user effort to reach the search target?
This study contributes to the research in this area by introducing a novel
ranking algorithm for type-based facets relying on a deep neural network. It uses
a combination of collaborative filtering (CF), query relevance, and personaliza-
tion features to rank the facets. The CF features reflect the general importance
of the t-facet among users. The query relevance derives the relevance of t-facets
to the query from the relevance of the resources to which they belong. Finally,
the personalization features exploit the user’s past preferences to build a vector
which represents the user interests. The extracted features are fed into a DNN
model. This is trained to combine these features into a t-facet ranking score. In
a following step, the approach provides a t-facet construction strategy to decide
the final tree to be portrayed to the searcher.
2 Facet Ranking Related Research
Several approaches have been proposed in literature to solve the problem of per-
sonalized facet ranking, which make use of individual user models, collaborative
filtering, or a mixture between the two. Factic [11] is a FSS that personalizes by
building user models from semantic usage logs. Several layers of user adaptation
are implemented and integrated with different weights to enhance the facet rel-
evance model. Koren et al. [9] suggested a CF approach by leveraging explicit
user feedback about the facets, which is used to build a facet relevance model for
individuals. They also use the aggregated ratings to build a collaborative model
for the new users in order to provide initial good facets in absence of a user
profile. A personalized ranking based on CF methods was suggested by Chanta-
munee et. al [5, 4]. They used user ratings and Matrix factorization with SVM
and Autoencoders to predict facet ranks. The Adaptive Twitter search system
generates user models from Twitter to personalize facet-values ordering [1]. The
user model contains entities extracted from the user’s tweets. The facet-values
are weighted higher if they exist in the user profile. Le et al. [10] also collect user
profiles from social networks. The profile is learned from user activities and pref-
erences using a TF-IDF feature vector model. Important facets are highlighted
through matching with the vector model. All the approaches discussed so far use
the same strategy to rank all types of facets. We believe it is important to distin-
guish between the types of facets during the ranking process as they support the
The benefits of buying summaries with Stuvia:
Guaranteed quality through customer reviews
Stuvia customers have reviewed more than 700,000 summaries. This how you know that you are buying the best documents.
Quick and easy check-out
You can quickly pay through credit card or Stuvia-credit for the summaries. There is no membership needed.
Focus on what matters
Your fellow students write the study notes themselves, which is why the documents are always reliable and up-to-date. This ensures you quickly get to the core!
Frequently asked questions
What do I get when I buy this document?
You get a PDF, available immediately after your purchase. The purchased document is accessible anytime, anywhere and indefinitely through your profile.
Satisfaction guarantee: how does it work?
Our satisfaction guarantee ensures that you always find a study document that suits you well. You fill out a form, and our customer service team takes care of the rest.
Who am I buying these notes from?
Stuvia is a marketplace, so you are not buying this document from us, but from seller Ariikelsey. Stuvia facilitates payment to the seller.
Will I be stuck with a subscription?
No, you only buy these notes for $14.99. You're not tied to anything after your purchase.
