@cataldomusto cataldo.musto@uniba.it
Justifying Recommendations through
Aspect-based Sentiment Analysis
of Users’ Reviews
CATALDO MUSTO, PASQUALE LOPS, MARCO DE GEMMIS, GIOVANNI SEMERARO
UNIVERSITÀ DEGLI STUDI DI BARI ‘ALDO MORO’ - ITALY
UMAP 2019 – 27th International
Conference on User Modeling,
Adaptation and Personalization
Larnaca (Cyprus) – June 10, 2019
Recommender
Systems
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Motivations
Recommendation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
I suggest you
Motivations
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
?
Recommendation
Intuition
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
To identify relevant and distinguishing
characteristics of the recommended
item by mining users’ reviews, and to use
these characteristics to build a natural
language justification
Intuition
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Breath-taking images
Excellent actors
Excellent sound design
Wonderful storytelling
To identify relevant and distinguishing
characteristics of the recommended
item by mining users’ reviews, and to use
these characteristics to build a natural
language justification
Intuition
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
"I recommend you 300 because people
who liked the movie think that the war
scenes are really well done. Moreover,
people liked 300 since it has a wonderful
storytelling and the actors are excellent"
A potential review-based justification
Solution
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Aspect Extraction
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Aspect Extraction
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify the aspects that are discussed when
people talk about the item
Strategy: to use natural language processing techniques
(specifically, part-of-speech tagging) to extract the names
mentioned in users’ reviews
Aspect Extraction
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify the aspects that are discussed when
people talk about the item
Strategy: to use natural language processing techniques
(specifically, part-of-speech tagging) to extract the names
mentioned in users’ reviews
Aspect Extraction
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
reviews aspects
Input: reviews of the item i R = {ri1, ri2 … rin}
Output: aspects A = {ai1, ai2 … aik}
Aspect Extraction
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
reviews aspects
Input: reviews of the item i R = {ri1, ri2 … rin}
Output: aspects A = {ai1, ai2 … aik}
Why only names?
Findings from previous
work in the area
Why no bigrams?
No significant
improvement emerged
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Aspect Extraction Goal: to identify the aspects that are
discussed when people talk about the item
Aspect Ranking Goal: to identify distinguishing aspects
that are discussed with a positive sentiment when people
talk about the item
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Aspect Extraction Goal: to identify the aspects that are
discussed when people talk about the item
Aspect Ranking Goal: to identify distinguishing aspects
that are discussed with a positive sentiment when people
talk about the item Strategy: to calculate a relevance score
for each aspect previously detected
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify distinguishing aspects that are discussed
with a positive sentiment when people talk about the item
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify distinguishing aspects that are discussed
with a positive sentiment when people talk about the item
How many times aspect ‘a’ appears in the
reviews of item ‘i’ (frequency of the aspect)
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify distinguishing aspects that are discussed
with a positive sentiment when people talk about the item
How many times aspect ‘a’ appears in the
reviews of item ‘i’ (frequency of the aspect)
How positive is the opinion of the users
when they talk about aspect ‘a’ (opinion
towards the aspect)
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify distinguishing aspects that are discussed
with a positive sentiment when people talk about the item
How many times aspect ‘a’ appears in the
reviews of item ‘i’ (frequency of the aspect)
How positive is the opinion of the users
when they talk about aspect ‘a’ (opinion
towards the aspect)
How distinguishing is
the aspect ‘a’ (inverse
popularity)
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to identify distinguishing aspects that are discussed
with a positive sentiment when people talk about the item
Intuition: our formula gives an higher score to the aspects that are
frequently mentioned in the reviews with a positive sentiment.
Moreover, it also rewards less popular aspects (higher IAF).
Aspect Ranking
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
aspects top-k aspects
Input: aspects A = {ai1, ai2 … aim}
Output: top-k aspects A = {ai1, ai2 … aik}
Generation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Generation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Goal: to generate a natural language justification that
relies on the most relevant aspects identified by the ASPECT
RANKING module.
For each aspect ’a’ returned by the ASPECT RANKING module
Browse the available reviews
Look for a compliant excerpt containing ‘a’
If the sentence has a positive sentiment
Add the sentence to the justification
Strategy
Generation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Question: when does an excerpt is a compliant sentence?
Answer: an excerpt is compliant if it follows one of the 18
justification patterns we defined
Example: the excerpt must have a third personal
singular verb
Generation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Question: when does an excerpt is a compliant sentence?
Answer: an excerpt is compliant if it follows one of the 18
justification patterns we defined
“I really liked the cast” Not compliant
“The cast was great” Compliant
Example: the excerpt must have a third personal
singular verb
Generation – Final Output
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
"I recommend you 300 because people who liked the
movie think that the war scenes are really well done.
Moreover, people liked 300 since it has a wonderful
storytelling and the actors are excellent"
Legenda
Blue: randomized sentences
Green: recommendation
Red: aspects (k=3)
Black: compliant excerpts
Experimental Evaluation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Research Question 1
How effective are the justifications generated through the pipeline, on varying of different
combinations of the parameters?
Research Question 2
How does our justifications perform with respect to a classic feature-based explanation [*] ?
(e.g.) “I suggest you 300 since you like other movies by Zack Snyder, as Watchmen”
Experimental Design
User Study with a Web Application
286 subjects
Movie and Books Domain
Metrics: Transparency, Engagement, Persuasion, Trust, Effectiveness [^]
Between-subjects for Research Question 1, Within-subjects for Research Question 2
[*] Musto, C., Narducci, F., Lops, P., De Gemmis, M., Semeraro G.
ExpLOD: A Framework for Explaining Recommendations based on
the Linked Open Data Cloud. In Proceedings of the 10th ACM
Conference on Recommender Systems. pp. 151-154. 2016
[^] Tintarev, N., & Masthoff, J. Designing and evaluating
explanations for recommender systems. In Recommender
systems handbook. pp. 479-510. Springer, Boston, MA. 2011
Experimental Evaluation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Parameters of the system
- Length of the justifications (short vs. long justifications)
short → top-2 aspects long -> top-4 aspects
- Vocabulary of aspects (statics vs. complete)
static → bounded to a fixed and pre-defined set of relevant aspects. No aspect
extraction, just aspect ranking
complete → not bounded. All the aspects are discovered by the aspect extractor
- Four different configurations
Implementation Details
Recommendations generated through Personalized PageRank, aspect extraction through
CoreNLP POS-tagger and sentiment analysis through Stanford algorithm
Experimental Evaluation
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Statistics of the datasets
Movies Books
#items 307 333
#reviews 153,566 52,560
avg. reviews/item 500.21 157.80
avg. words/reviews 138.38 126.65
Dataset are obtained by mapping items from MovieLens and BookCrossing
datasets with reviews obtained from Amazon data (dataset available)
Web Application
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Recommendation
Web Application (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Recommendation
Justification
Questionnaire
Web Application (Research Question 2)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
I propose you “Aliens”
because you sometimes like
movies edited by Canadian
film editors, American fiction
films and 1980s films, as The
Terminator.
I recommend you “Aliens”
because people who liked this
movie think that the Alien
series is one of the best sci-fi
movies and that the ending is
awesome with some fantastic
action scenes.
Review-based
Explanation
Feature-based
Explanation
Results (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
MOVIES Transparency Persuasion Engagement Trust Effectiveness
Static Short 3.40 3.13 3.09 3.23 0.64
Static Long 3.77 3.68 3.55 3.73 0.55
Complete Short 3.91 3.60 3.25 3.70 0.53
Complete Long 3.74 3.48 3.35 3.46 0.59
Results (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Finding 1
With the ‘complete’ set of
aspects, shorter justifications
have the best results
MOVIES Transparency Persuasion Engagement Trust Effectiveness
Static Short 3.40 3.13 3.09 3.23 0.64
Static Long 3.77 3.68 3.55 3.73 0.55
Complete Short 3.91 3.60 3.25 3.70 0.53
Complete Long 3.74 3.48 3.35 3.46 0.59
Results (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Finding 1
With the ‘complete’ set of
aspects, shorter justifications
have the best results
Finding 2
With the ‘static’ set of
aspects, longer justifications
have the best results
MOVIES Transparency Persuasion Engagement Trust Effectiveness
Static Short 3.40 3.13 3.09 3.23 0.64
Static Long 3.77 3.68 3.55 3.73 0.55
Complete Short 3.91 3.60 3.25 3.70 0.53
Complete Long 3.74 3.48 3.35 3.46 0.59
Results (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Finding 1
With the ‘complete’ set of
aspects, shorter justifications
have the best results
Finding 2
With the ‘static’ set of
aspects, longer justifications
have the best results
Overall
Long justifications based on
static aspects have the best
results in the Movie Domain
MOVIES Transparency Persuasion Engagement Trust Effectiveness
Static Short 3.40 3.13 3.09 3.23 0.64
Static Long 3.77 3.68 3.55 3.73 0.55
Complete Short 3.91 3.60 3.25 3.70 0.53
Complete Long 3.74 3.48 3.35 3.46 0.59
Results (Research Question 1)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
BOOKS Transparency Persuasion Engagement Trust Effectiveness
Static Short 3.57 3.40 3.07 3.31 0.74
Static Long 3.64 3.82 3.45 3.71 0.45
Complete Short 3.82 3.59 3.37 3.51 0.47
Complete Long 3.61 3.41 3.31 3.30 0.57
Findings confirmed for the Books domain.
Static list of aspects + Long justifications = Best Performance
Results (Research Question 2)
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
MOVIES
Review
-based
Feature-
based
Indifferent
Transparency 47.4% 38.6% 14.0%
Persuasion 51.7% 43.3% 5.0%
Engagement 66.7% 25.0% 8.3%
Trust 53.3% 35.5% 11.7%
Effectiveness 57.9% 35.0% 7.1%
Outcome: Users preferred Review-based Justifications
Confirmed for all the metrics and both the domains
BOOKS
Review-
based
Feature-
based
Indifferent
Transparency 58.1% 36.0% 5.9%
Persuasion 61.8% 29.0% 9.2%
Engagement 54.6% 27.3% 18.1%
Trust 58.2% 27.2% 14.6%
Effectiveness 59.9% 31.1% 10.0%
Recap
Exploitation of Users’ Review to Generate Natural Language Justification
Natural Language Processing and Sentiment Analysis pipeline
◦ Aspect Extraction
◦ Aspect Ranking
◦ Generation of the Justification
Outcomes of the Experiments
◦ If we bound the generation to a static list of aspects, we can generate longer justifications
◦ Otherwise shorter justifications are preferred (due to some noise?)
◦ In general, review-based justifications are preferred to classic feature-based explanations
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Future Work
To evaluate new and different domains
To improve the implementation of the single modules
◦ Aspect Extraction (try to extract bigrams or entities)
◦ Aspect Ranking (more sophisticated techniques can be exploited)
◦ Generation of the Justification (make the natural language generation more dynamic)
To introduce methods to generate hybrid explanations
◦ Combinations of review data with descriptive features
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
Thank you!
cataldo.musto@uniba.it
@cataldomusto
Contacts
Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro.
Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019

Justifying Recommendations through Aspect-based Sentiment Analysis of Users Reviews

  • 1.
    @cataldomusto cataldo.musto@uniba.it Justifying Recommendationsthrough Aspect-based Sentiment Analysis of Users’ Reviews CATALDO MUSTO, PASQUALE LOPS, MARCO DE GEMMIS, GIOVANNI SEMERARO UNIVERSITÀ DEGLI STUDI DI BARI ‘ALDO MORO’ - ITALY UMAP 2019 – 27th International Conference on User Modeling, Adaptation and Personalization Larnaca (Cyprus) – June 10, 2019
  • 2.
    Recommender Systems Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 3.
    Motivations Recommendation Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 I suggest you
  • 4.
    Motivations Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 ? Recommendation
  • 5.
    Intuition Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 To identify relevant and distinguishing characteristics of the recommended item by mining users’ reviews, and to use these characteristics to build a natural language justification
  • 6.
    Intuition Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Breath-taking images Excellent actors Excellent sound design Wonderful storytelling To identify relevant and distinguishing characteristics of the recommended item by mining users’ reviews, and to use these characteristics to build a natural language justification
  • 7.
    Intuition Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 "I recommend you 300 because people who liked the movie think that the war scenes are really well done. Moreover, people liked 300 since it has a wonderful storytelling and the actors are excellent" A potential review-based justification
  • 8.
    Solution Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 9.
    Aspect Extraction Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 10.
    Aspect Extraction Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify the aspects that are discussed when people talk about the item Strategy: to use natural language processing techniques (specifically, part-of-speech tagging) to extract the names mentioned in users’ reviews
  • 11.
    Aspect Extraction Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify the aspects that are discussed when people talk about the item Strategy: to use natural language processing techniques (specifically, part-of-speech tagging) to extract the names mentioned in users’ reviews
  • 12.
    Aspect Extraction Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 reviews aspects Input: reviews of the item i R = {ri1, ri2 … rin} Output: aspects A = {ai1, ai2 … aik}
  • 13.
    Aspect Extraction Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 reviews aspects Input: reviews of the item i R = {ri1, ri2 … rin} Output: aspects A = {ai1, ai2 … aik} Why only names? Findings from previous work in the area Why no bigrams? No significant improvement emerged
  • 14.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 15.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Aspect Extraction Goal: to identify the aspects that are discussed when people talk about the item Aspect Ranking Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item
  • 16.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Aspect Extraction Goal: to identify the aspects that are discussed when people talk about the item Aspect Ranking Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item Strategy: to calculate a relevance score for each aspect previously detected
  • 17.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item
  • 18.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item How many times aspect ‘a’ appears in the reviews of item ‘i’ (frequency of the aspect)
  • 19.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item How many times aspect ‘a’ appears in the reviews of item ‘i’ (frequency of the aspect) How positive is the opinion of the users when they talk about aspect ‘a’ (opinion towards the aspect)
  • 20.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item How many times aspect ‘a’ appears in the reviews of item ‘i’ (frequency of the aspect) How positive is the opinion of the users when they talk about aspect ‘a’ (opinion towards the aspect) How distinguishing is the aspect ‘a’ (inverse popularity)
  • 21.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to identify distinguishing aspects that are discussed with a positive sentiment when people talk about the item Intuition: our formula gives an higher score to the aspects that are frequently mentioned in the reviews with a positive sentiment. Moreover, it also rewards less popular aspects (higher IAF).
  • 22.
    Aspect Ranking Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 aspects top-k aspects Input: aspects A = {ai1, ai2 … aim} Output: top-k aspects A = {ai1, ai2 … aik}
  • 23.
    Generation Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 24.
    Generation Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Goal: to generate a natural language justification that relies on the most relevant aspects identified by the ASPECT RANKING module. For each aspect ’a’ returned by the ASPECT RANKING module Browse the available reviews Look for a compliant excerpt containing ‘a’ If the sentence has a positive sentiment Add the sentence to the justification Strategy
  • 25.
    Generation Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Question: when does an excerpt is a compliant sentence? Answer: an excerpt is compliant if it follows one of the 18 justification patterns we defined Example: the excerpt must have a third personal singular verb
  • 26.
    Generation Cataldo Musto, PasqualeLops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Question: when does an excerpt is a compliant sentence? Answer: an excerpt is compliant if it follows one of the 18 justification patterns we defined “I really liked the cast” Not compliant “The cast was great” Compliant Example: the excerpt must have a third personal singular verb
  • 27.
    Generation – FinalOutput Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 "I recommend you 300 because people who liked the movie think that the war scenes are really well done. Moreover, people liked 300 since it has a wonderful storytelling and the actors are excellent" Legenda Blue: randomized sentences Green: recommendation Red: aspects (k=3) Black: compliant excerpts
  • 28.
    Experimental Evaluation Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Research Question 1 How effective are the justifications generated through the pipeline, on varying of different combinations of the parameters? Research Question 2 How does our justifications perform with respect to a classic feature-based explanation [*] ? (e.g.) “I suggest you 300 since you like other movies by Zack Snyder, as Watchmen” Experimental Design User Study with a Web Application 286 subjects Movie and Books Domain Metrics: Transparency, Engagement, Persuasion, Trust, Effectiveness [^] Between-subjects for Research Question 1, Within-subjects for Research Question 2 [*] Musto, C., Narducci, F., Lops, P., De Gemmis, M., Semeraro G. ExpLOD: A Framework for Explaining Recommendations based on the Linked Open Data Cloud. In Proceedings of the 10th ACM Conference on Recommender Systems. pp. 151-154. 2016 [^] Tintarev, N., & Masthoff, J. Designing and evaluating explanations for recommender systems. In Recommender systems handbook. pp. 479-510. Springer, Boston, MA. 2011
  • 29.
    Experimental Evaluation Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Parameters of the system - Length of the justifications (short vs. long justifications) short → top-2 aspects long -> top-4 aspects - Vocabulary of aspects (statics vs. complete) static → bounded to a fixed and pre-defined set of relevant aspects. No aspect extraction, just aspect ranking complete → not bounded. All the aspects are discovered by the aspect extractor - Four different configurations Implementation Details Recommendations generated through Personalized PageRank, aspect extraction through CoreNLP POS-tagger and sentiment analysis through Stanford algorithm
  • 30.
    Experimental Evaluation Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Statistics of the datasets Movies Books #items 307 333 #reviews 153,566 52,560 avg. reviews/item 500.21 157.80 avg. words/reviews 138.38 126.65 Dataset are obtained by mapping items from MovieLens and BookCrossing datasets with reviews obtained from Amazon data (dataset available)
  • 31.
    Web Application Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Recommendation
  • 32.
    Web Application (ResearchQuestion 1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Recommendation Justification Questionnaire
  • 33.
    Web Application (ResearchQuestion 2) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 I propose you “Aliens” because you sometimes like movies edited by Canadian film editors, American fiction films and 1980s films, as The Terminator. I recommend you “Aliens” because people who liked this movie think that the Alien series is one of the best sci-fi movies and that the ending is awesome with some fantastic action scenes. Review-based Explanation Feature-based Explanation
  • 34.
    Results (Research Question1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 MOVIES Transparency Persuasion Engagement Trust Effectiveness Static Short 3.40 3.13 3.09 3.23 0.64 Static Long 3.77 3.68 3.55 3.73 0.55 Complete Short 3.91 3.60 3.25 3.70 0.53 Complete Long 3.74 3.48 3.35 3.46 0.59
  • 35.
    Results (Research Question1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Finding 1 With the ‘complete’ set of aspects, shorter justifications have the best results MOVIES Transparency Persuasion Engagement Trust Effectiveness Static Short 3.40 3.13 3.09 3.23 0.64 Static Long 3.77 3.68 3.55 3.73 0.55 Complete Short 3.91 3.60 3.25 3.70 0.53 Complete Long 3.74 3.48 3.35 3.46 0.59
  • 36.
    Results (Research Question1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Finding 1 With the ‘complete’ set of aspects, shorter justifications have the best results Finding 2 With the ‘static’ set of aspects, longer justifications have the best results MOVIES Transparency Persuasion Engagement Trust Effectiveness Static Short 3.40 3.13 3.09 3.23 0.64 Static Long 3.77 3.68 3.55 3.73 0.55 Complete Short 3.91 3.60 3.25 3.70 0.53 Complete Long 3.74 3.48 3.35 3.46 0.59
  • 37.
    Results (Research Question1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 Finding 1 With the ‘complete’ set of aspects, shorter justifications have the best results Finding 2 With the ‘static’ set of aspects, longer justifications have the best results Overall Long justifications based on static aspects have the best results in the Movie Domain MOVIES Transparency Persuasion Engagement Trust Effectiveness Static Short 3.40 3.13 3.09 3.23 0.64 Static Long 3.77 3.68 3.55 3.73 0.55 Complete Short 3.91 3.60 3.25 3.70 0.53 Complete Long 3.74 3.48 3.35 3.46 0.59
  • 38.
    Results (Research Question1) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 BOOKS Transparency Persuasion Engagement Trust Effectiveness Static Short 3.57 3.40 3.07 3.31 0.74 Static Long 3.64 3.82 3.45 3.71 0.45 Complete Short 3.82 3.59 3.37 3.51 0.47 Complete Long 3.61 3.41 3.31 3.30 0.57 Findings confirmed for the Books domain. Static list of aspects + Long justifications = Best Performance
  • 39.
    Results (Research Question2) Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019 MOVIES Review -based Feature- based Indifferent Transparency 47.4% 38.6% 14.0% Persuasion 51.7% 43.3% 5.0% Engagement 66.7% 25.0% 8.3% Trust 53.3% 35.5% 11.7% Effectiveness 57.9% 35.0% 7.1% Outcome: Users preferred Review-based Justifications Confirmed for all the metrics and both the domains BOOKS Review- based Feature- based Indifferent Transparency 58.1% 36.0% 5.9% Persuasion 61.8% 29.0% 9.2% Engagement 54.6% 27.3% 18.1% Trust 58.2% 27.2% 14.6% Effectiveness 59.9% 31.1% 10.0%
  • 40.
    Recap Exploitation of Users’Review to Generate Natural Language Justification Natural Language Processing and Sentiment Analysis pipeline ◦ Aspect Extraction ◦ Aspect Ranking ◦ Generation of the Justification Outcomes of the Experiments ◦ If we bound the generation to a static list of aspects, we can generate longer justifications ◦ Otherwise shorter justifications are preferred (due to some noise?) ◦ In general, review-based justifications are preferred to classic feature-based explanations Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 41.
    Future Work To evaluatenew and different domains To improve the implementation of the single modules ◦ Aspect Extraction (try to extract bigrams or entities) ◦ Aspect Ranking (more sophisticated techniques can be exploited) ◦ Generation of the Justification (make the natural language generation more dynamic) To introduce methods to generate hybrid explanations ◦ Combinations of review data with descriptive features Cataldo Musto, Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019
  • 42.
    Thank you! cataldo.musto@uniba.it @cataldomusto Contacts Cataldo Musto,Pasquale Lops, Marco de Gemmis, Giovanni Semeraro. Justifying Recommendations through Aspect-based Sentiment Analysis of Users’ Reviews. UMAP 2019 – Larnaca (Cyprus) – June 10, 2019