1) The document discusses using a Rasch model to create a psychometric user model for recommending energy conservation measures and lifestyle changes for hypertension management.
2) A Rasch scale was developed to represent individual abilities and difficulties of different conservation measures and health behaviors along a single dimension.
3) Tailoring recommendations based on individuals' position on the Rasch scale was found to increase perceived support, choice satisfaction, and likelihood of choosing recommended items compared to random recommendations.
Enhancement of the Neutrality in Recommendation
Workshop on Human Decision Making in Recommender Systems, in conjunction with RecSys 2012
Article @ Official Site: http://ceur-ws.org/Vol-893/
Article @ Personal Site: http://www.kamishima.net/archive/2012-ws-recsys-print.pdf
Handnote : http://www.kamishima.net/archive/2012-ws-recsys-HN.pdf
Program codes : http://www.kamishima.net/inrs
Workshop Homepage: http://recex.ist.tugraz.at/RecSysWorkshop2012
Abstract:
This paper proposes an algorithm for making recommendation so that the neutrality toward the viewpoint specified by a user is enhanced. This algorithm is useful for avoiding to make decisions based on biased information. Such a problem is pointed out as the filter bubble, which is the influence in social decisions biased by a personalization technology. To provide such a recommendation, we assume that a user specifies a viewpoint toward which the user want to enforce the neutrality, because recommendation that is neutral from any information is no longer recommendation. Given such a target viewpoint, we implemented information neutral recommendation algorithm by introducing a penalty term to enforce the statistical independence between the target viewpoint and a preference score. We empirically show that our algorithm enhances the independence toward the specified viewpoint by and then demonstrate how sets of recommended items are changed.
Efficiency Improvement of Neutrality-Enhanced RecommendationToshihiro Kamishima
Efficiency Improvement of Neutrality-Enhanced Recommendation
Workshop on Human Decision Making in Recommender Systems, in conjunction with RecSys 2013
Article @ Official Site: http://ceur-ws.org/Vol-1050/
Article @ Personal Site: http://www.kamishima.net/archive/2013-ws-recsys-print.pdf
Handnote : http://www.kamishima.net/archive/2013-ws-recsys-HN.pdf
Program codes : http://www.kamishima.net/inrs/
Workshop Homepage: http://recex.ist.tugraz.at/RecSysWorkshop/
Abstract:
This paper proposes an algorithm for making recommendations so that neutrality from a viewpoint specified by the user is enhanced. This algorithm is useful for avoiding decisions based on biased information. Such a problem is pointed out as the filter bubble, which is the influence in social decisions biased by personalization technologies. To provide a neutrality-enhanced recommendation, we must first assume that a user can specify a particular viewpoint from which the neutrality can be applied, because a recommendation that is neutral from all viewpoints is no longer a recommendation. Given such a target viewpoint, we implement an information-neutral recommendation algorithm by introducing a penalty term to enforce statistical independence between the target viewpoint and a rating. We empirically show that our algorithm enhances the independence from the specified viewpoint.
(Gaurav sawant & dhaval sawlani)bia 678 final project reportGaurav Sawant
PROJECT REPORT
• Performed memory-based collaborative filtering techniques like Cosine similarities, Pearson’s r & model-based Matrix Factorization techniques like Alternating Least Squares (ALS) method
• Studied the scalability of these methods on local machines & on Hadoop clusters
On building more human query answering systemsINRIA-OAK
The underlying principle behind every query answering system is the existence of a query describing the information of interest. When this model is applied to non-expert users, two traditional issues become highly significant.
The first is that many queries are often over specified leading to empty answers. We propose a principled optimization-based interactive query relaxation framework for such queries. The framework computes dynamically and suggests alternative queries with less conditions to help the user arrive at a query with a non-empty answer, or at a query for which it is clear that independently of the relaxations the answer will always be empty.
The second issue is the lack of expertise from the user to accurately describe the requirements of the elements of interest. The user may though know examples of elements that would like to have in the results. We introduce a novel form of query paradigm in which queries are not any more specifications of what the user is searching for, but simply a sample of what the user knows to be of interest. We refer to this novel form of queries as Exemplar Queries.
Enhancement of the Neutrality in Recommendation
Workshop on Human Decision Making in Recommender Systems, in conjunction with RecSys 2012
Article @ Official Site: http://ceur-ws.org/Vol-893/
Article @ Personal Site: http://www.kamishima.net/archive/2012-ws-recsys-print.pdf
Handnote : http://www.kamishima.net/archive/2012-ws-recsys-HN.pdf
Program codes : http://www.kamishima.net/inrs
Workshop Homepage: http://recex.ist.tugraz.at/RecSysWorkshop2012
Abstract:
This paper proposes an algorithm for making recommendation so that the neutrality toward the viewpoint specified by a user is enhanced. This algorithm is useful for avoiding to make decisions based on biased information. Such a problem is pointed out as the filter bubble, which is the influence in social decisions biased by a personalization technology. To provide such a recommendation, we assume that a user specifies a viewpoint toward which the user want to enforce the neutrality, because recommendation that is neutral from any information is no longer recommendation. Given such a target viewpoint, we implemented information neutral recommendation algorithm by introducing a penalty term to enforce the statistical independence between the target viewpoint and a preference score. We empirically show that our algorithm enhances the independence toward the specified viewpoint by and then demonstrate how sets of recommended items are changed.
Efficiency Improvement of Neutrality-Enhanced RecommendationToshihiro Kamishima
Efficiency Improvement of Neutrality-Enhanced Recommendation
Workshop on Human Decision Making in Recommender Systems, in conjunction with RecSys 2013
Article @ Official Site: http://ceur-ws.org/Vol-1050/
Article @ Personal Site: http://www.kamishima.net/archive/2013-ws-recsys-print.pdf
Handnote : http://www.kamishima.net/archive/2013-ws-recsys-HN.pdf
Program codes : http://www.kamishima.net/inrs/
Workshop Homepage: http://recex.ist.tugraz.at/RecSysWorkshop/
Abstract:
This paper proposes an algorithm for making recommendations so that neutrality from a viewpoint specified by the user is enhanced. This algorithm is useful for avoiding decisions based on biased information. Such a problem is pointed out as the filter bubble, which is the influence in social decisions biased by personalization technologies. To provide a neutrality-enhanced recommendation, we must first assume that a user can specify a particular viewpoint from which the neutrality can be applied, because a recommendation that is neutral from all viewpoints is no longer a recommendation. Given such a target viewpoint, we implement an information-neutral recommendation algorithm by introducing a penalty term to enforce statistical independence between the target viewpoint and a rating. We empirically show that our algorithm enhances the independence from the specified viewpoint.
(Gaurav sawant & dhaval sawlani)bia 678 final project reportGaurav Sawant
PROJECT REPORT
• Performed memory-based collaborative filtering techniques like Cosine similarities, Pearson’s r & model-based Matrix Factorization techniques like Alternating Least Squares (ALS) method
• Studied the scalability of these methods on local machines & on Hadoop clusters
On building more human query answering systemsINRIA-OAK
The underlying principle behind every query answering system is the existence of a query describing the information of interest. When this model is applied to non-expert users, two traditional issues become highly significant.
The first is that many queries are often over specified leading to empty answers. We propose a principled optimization-based interactive query relaxation framework for such queries. The framework computes dynamically and suggests alternative queries with less conditions to help the user arrive at a query with a non-empty answer, or at a query for which it is clear that independently of the relaxations the answer will always be empty.
The second issue is the lack of expertise from the user to accurately describe the requirements of the elements of interest. The user may though know examples of elements that would like to have in the results. We introduce a novel form of query paradigm in which queries are not any more specifications of what the user is searching for, but simply a sample of what the user knows to be of interest. We refer to this novel form of queries as Exemplar Queries.
A REVIEW PAPER ON BFO AND PSO BASED MOVIE RECOMMENDATION SYSTEM | J4RV4I1015Journal For Research
Recommendation system plays important role in Internet world and used in many applications. It has created the collection of many application, created global village and growth for numerous information. This paper represents the overview of Approaches and techniques generated in recommendation system. Recommendation system is categorized in three classes: Collaborative Filtering, Content based and hybrid based Approach. This paper classifies collaborative filtering in two types: Memory based and Model based Recommendation .The paper elaborates these approaches and their techniques with their limitations. The result of our system provides much better recommendations to users because it enables the users to understand the relation between their emotional states and the recommended movies.
Fairness-aware Learning through Regularization Approach
The 3rd IEEE International Workshop on Privacy Aspects of Data Mining (PADM 2011)
Dec. 11, 2011 @ Vancouver, Canada, in conjunction with ICDM2011
Article @ Official Site: http://doi.ieeecomputersociety.org/10.1109/ICDMW.2011.83
Article @ Personal Site: http://www.kamishima.net/archive/2011-ws-icdm_padm.pdf
Handnote: http://www.kamishima.net/archive/2011-ws-icdm_padm-HN.pdf
Workshop Homepage: http://www.zurich.ibm.com/padm2011/
Abstract:
With the spread of data mining technologies and the accumulation of social data, such technologies and data are being used for determinations that seriously affect people's lives. For example, credit scoring is frequently determined based on the records of past credit data together with statistical prediction techniques. Needless to say, such determinations must be socially and legally fair from a viewpoint of social responsibility; namely, it must be unbiased and nondiscriminatory in sensitive features, such as race, gender, religion, and so on. Several researchers have recently begun to attempt the development of analysis techniques that are aware of social fairness or discrimination. They have shown that simply avoiding the use of sensitive features is insufficient for eliminating biases in determinations, due to the indirect influence of sensitive information. From a privacy-preserving viewpoint, this can be interpreted as hiding sensitive information when classification results are observed. In this paper, we first discuss three causes of unfairness in machine learning. We then propose a regularization approach that is applicable to any prediction algorithm with probabilistic discriminative models. We further apply this approach to logistic regression and empirically show its effectiveness and efficiency.
Consideration on Fairness-aware Data Mining
IEEE International Workshop on Discrimination and Privacy-Aware Data Mining (DPADM 2012)
Dec. 10, 2012 @ Brussels, Belgium, in conjunction with ICDM2012
Article @ Official Site: http://doi.ieeecomputersociety.org/10.1109/ICDMW.2012.101
Article @ Personal Site: http://www.kamishima.net/archive/2012-ws-icdm-print.pdf
Handnote: http://www.kamishima.net/archive/2012-ws-icdm-HN.pdf
Workshop Homepage: https://sites.google.com/site/dpadm2012/
Abstract:
With the spread of data mining technologies and the accumulation of social data, such technologies and data are being used for determinations that seriously affect individuals' lives. For example, credit scoring is frequently determined based on the records of past credit data together with statistical prediction techniques. Needless to say, such determinations must be nondiscriminatory and fair regarding sensitive features such as race, gender, religion, and so on. Several researchers have recently begun to develop fairness-aware or discrimination-aware data mining techniques that take into account issues of social fairness, discrimination, and neutrality. In this paper, after demonstrating the applications of these techniques, we explore the formal concepts of fairness and techniques for handling fairness in data mining. We then provide an integrated view of these concepts based on statistical independence. Finally, we discuss the relations between fairness-aware data mining and other research topics, such as privacy-preserving data mining or causal inference.
2 Studies UX types should know about (Straub UXPA unconference13)Kath Straub
I described these two studies during the Research in Practice: Studies UXers should know about workshop. I expected them to be drive-bys ... as in, "Yah, yah, .. have heard that ... let's move on." I was surprised to find that the group -- a sharp, engaged and thoughtful group-- didn't know these studies. Instead of a few minutes description, we discussed and debated how these studies might influence UX practice for almost an hour. Based on that, I got nudged (Culprit = @susandra Susan Dray) to presenting these two @ the UXPA unconference.
There are many other studies studies that all UXPros should be familiar with ...
A REVIEW PAPER ON BFO AND PSO BASED MOVIE RECOMMENDATION SYSTEM | J4RV4I1015Journal For Research
Recommendation system plays important role in Internet world and used in many applications. It has created the collection of many application, created global village and growth for numerous information. This paper represents the overview of Approaches and techniques generated in recommendation system. Recommendation system is categorized in three classes: Collaborative Filtering, Content based and hybrid based Approach. This paper classifies collaborative filtering in two types: Memory based and Model based Recommendation .The paper elaborates these approaches and their techniques with their limitations. The result of our system provides much better recommendations to users because it enables the users to understand the relation between their emotional states and the recommended movies.
Fairness-aware Learning through Regularization Approach
The 3rd IEEE International Workshop on Privacy Aspects of Data Mining (PADM 2011)
Dec. 11, 2011 @ Vancouver, Canada, in conjunction with ICDM2011
Article @ Official Site: http://doi.ieeecomputersociety.org/10.1109/ICDMW.2011.83
Article @ Personal Site: http://www.kamishima.net/archive/2011-ws-icdm_padm.pdf
Handnote: http://www.kamishima.net/archive/2011-ws-icdm_padm-HN.pdf
Workshop Homepage: http://www.zurich.ibm.com/padm2011/
Abstract:
With the spread of data mining technologies and the accumulation of social data, such technologies and data are being used for determinations that seriously affect people's lives. For example, credit scoring is frequently determined based on the records of past credit data together with statistical prediction techniques. Needless to say, such determinations must be socially and legally fair from a viewpoint of social responsibility; namely, it must be unbiased and nondiscriminatory in sensitive features, such as race, gender, religion, and so on. Several researchers have recently begun to attempt the development of analysis techniques that are aware of social fairness or discrimination. They have shown that simply avoiding the use of sensitive features is insufficient for eliminating biases in determinations, due to the indirect influence of sensitive information. From a privacy-preserving viewpoint, this can be interpreted as hiding sensitive information when classification results are observed. In this paper, we first discuss three causes of unfairness in machine learning. We then propose a regularization approach that is applicable to any prediction algorithm with probabilistic discriminative models. We further apply this approach to logistic regression and empirically show its effectiveness and efficiency.
Consideration on Fairness-aware Data Mining
IEEE International Workshop on Discrimination and Privacy-Aware Data Mining (DPADM 2012)
Dec. 10, 2012 @ Brussels, Belgium, in conjunction with ICDM2012
Article @ Official Site: http://doi.ieeecomputersociety.org/10.1109/ICDMW.2012.101
Article @ Personal Site: http://www.kamishima.net/archive/2012-ws-icdm-print.pdf
Handnote: http://www.kamishima.net/archive/2012-ws-icdm-HN.pdf
Workshop Homepage: https://sites.google.com/site/dpadm2012/
Abstract:
With the spread of data mining technologies and the accumulation of social data, such technologies and data are being used for determinations that seriously affect individuals' lives. For example, credit scoring is frequently determined based on the records of past credit data together with statistical prediction techniques. Needless to say, such determinations must be nondiscriminatory and fair regarding sensitive features such as race, gender, religion, and so on. Several researchers have recently begun to develop fairness-aware or discrimination-aware data mining techniques that take into account issues of social fairness, discrimination, and neutrality. In this paper, after demonstrating the applications of these techniques, we explore the formal concepts of fairness and techniques for handling fairness in data mining. We then provide an integrated view of these concepts based on statistical independence. Finally, we discuss the relations between fairness-aware data mining and other research topics, such as privacy-preserving data mining or causal inference.
2 Studies UX types should know about (Straub UXPA unconference13)Kath Straub
I described these two studies during the Research in Practice: Studies UXers should know about workshop. I expected them to be drive-bys ... as in, "Yah, yah, .. have heard that ... let's move on." I was surprised to find that the group -- a sharp, engaged and thoughtful group-- didn't know these studies. Instead of a few minutes description, we discussed and debated how these studies might influence UX practice for almost an hour. Based on that, I got nudged (Culprit = @susandra Susan Dray) to presenting these two @ the UXPA unconference.
There are many other studies studies that all UXPros should be familiar with ...
Fairness in Search & RecSys 네이버 검색 콜로키움 김진영Jin Young Kim
검색 및 추천 시스템의 사회적 역할이 커지면서, 그 결과의 공정성 역시 최근 관심사로 대두되었다. 본 발표에서는 검색 및 추천시스템의 공정성 이슈 및 그 해법을 다룬다. 공정한 검색 및 추천 결과를 정의하는 다양한 방법, 공정성의 결여가 미치는 자원 배분 및 스테레오타이핑 문제, 그리고 검색 및 추천시스템 개발의 각 단계별로 어떤 해결책이 있는지를 최신 연구 중심으로 살펴본다. 마지막으로 실제 공정한 시스템 개발을 위한 실무적인 고려사항을 다룬다.
The International Journal of Engineering and Science (The IJES)theijes
The International Journal of Engineering & Science is aimed at providing a platform for researchers, engineers, scientists, or educators to publish their original research results, to exchange new ideas, to disseminate information in innovative designs, engineering experiences and technological skills. It is also the Journal's objective to promote engineering and technology education. All papers submitted to the Journal will be blind peer-reviewed. Only original articles will be published.
A recommender system(RS) is an information filtering system that recommends the related suggestions as per the end users requirement. Applications of RS include recommendation of movies, music, serials, books, documents, websites, tourist places etc.
Benefits of RS: RSs are beneficial to both service providers and to the users. RSs reduce transaction costs of finding and selecting items.& RSs help in decision making. The proposed work DEMOGRAPHY BASED HYBRID SYSTEM FOR MOVIE RECOMMENDATIONS highlights the combination of collaborative, content based & demographic filtering to recommend movies to the end user. The model uses SVD++ technique available in Surprise Python library and achieves the MSE of 0.92 which is comparatively less than the other techniques.
Recommender Systems Fairness Evaluation via Generalized Cross EntropyVito Walter Anelli
Fairness in recommender systems has been considered with respect to sensitive attributes of users (e.g., gender, race) or items (e.g., revenue in a multistakeholder setting). Regardless, the concept has been commonly interpreted as some form of equality – i.e., the degree to which the system is meeting the information needs of all its users in an equal sense. In this paper, we argue that fairness in recommender systems does not necessarily imply equality, but instead it should consider a distribution of resources based on merits and needs. We present a probabilistic framework based on generalized cross entropy to evaluate fairness of recommender systems under this perspective, where we show that the proposed framework is flexible and explanatory by allowing to incorporate domain knowledge (through an ideal fair distribution) that can help to understand which item or user aspects a recommendation algorithm is over- or under-representing. Results on two real-world datasets show the merits of the proposed evaluation framework both in terms of user and item fairness.
This presentation by Morris Kleiner (University of Minnesota), was made during the discussion “Competition and Regulation in Professions and Occupations” held at the Working Party No. 2 on Competition and Regulation on 10 June 2024. More papers and presentations on the topic can be found out at oe.cd/crps.
This presentation was uploaded with the author’s consent.
Acorn Recovery: Restore IT infra within minutesIP ServerOne
Introducing Acorn Recovery as a Service, a simple, fast, and secure managed disaster recovery (DRaaS) by IP ServerOne. A DR solution that helps restore your IT infra within minutes.
Sharpen existing tools or get a new toolbox? Contemporary cluster initiatives...Orkestra
UIIN Conference, Madrid, 27-29 May 2024
James Wilson, Orkestra and Deusto Business School
Emily Wise, Lund University
Madeline Smith, The Glasgow School of Art
0x01 - Newton's Third Law: Static vs. Dynamic AbusersOWASP Beja
f you offer a service on the web, odds are that someone will abuse it. Be it an API, a SaaS, a PaaS, or even a static website, someone somewhere will try to figure out a way to use it to their own needs. In this talk we'll compare measures that are effective against static attackers and how to battle a dynamic attacker who adapts to your counter-measures.
About the Speaker
===============
Diogo Sousa, Engineering Manager @ Canonical
An opinionated individual with an interest in cryptography and its intersection with secure software development.
Have you ever wondered how search works while visiting an e-commerce site, internal website, or searching through other types of online resources? Look no further than this informative session on the ways that taxonomies help end-users navigate the internet! Hear from taxonomists and other information professionals who have first-hand experience creating and working with taxonomies that aid in navigation, search, and discovery across a range of disciplines.
Announcement of 18th IEEE International Conference on Software Testing, Verif...
What recommender systems can learn from decision psychology about preference elicitation and behavioral change
1. What recommender
systems can learn from
decision psychology
about preference elicitation
and behavioral change
Martijn Willemsen
Human-Technology Interaction
www.martijnwillemsen.nl
2. What are recommender systems about
Algorithms
Accuracy:
compare prediction
with actual values
Recommendation:
best predicted items
dataset
user-item rating pairs
user
Choose (prefer?)
ratings
Rating?
Experience!
preferences
Goals &
desires!
6. User-Centric Framework
Our framework adds the intermediate construct of perception that explains
why behavior and experiences changes due to our manipulations
7. User-Centric Framework
And adds personal
and situational
characteristics
Relations modeled
using factor analysis
and SEM
Knijnenburg, B.P., Willemsen, M.C., Gantner, Z., Soncu, H., Newell, C. (2012). Explaining
the User Experience of Recommender Systems. User Modeling and User-Adapted
Interaction (UMUAI), vol 22, p. 441-504 http://bit.ly/umuai
8. Providing input to the
recommender:
Preference Elicitation
memory, ratings & choices
9. Providing input to a recommender
system
how do algorithms get their data?
Preference Elicitation (PE)
PE is a major topic in research on
Decision Making
I even did my PhD thesis on it… ;-)
What can Psychology learn us on
improving this aspect?
Role of memory in ratings
Rating support
Rating versus choice-based elicitation
10. What does rating entail?
Typical recommender scenario:
First usage: Show a set of (typical) items and ask
people to rate the ones they know (cold start)
Later usage: people go to the recommender when they
have consumed an item to rate it and typically also rate
other aspects
Does it matter if the preference you provide (by
rating) is based on recent experiences or mostly
on your memory?
11. Psychologist: user knows best, ask her!
In two user experiments, users rated a number of
movies that were aired on Dutch TV in the
previous month (~150 movies)
Rate 15-20 movies from that set that you know
and indicate how long ago you have seen these
movies (last week, last month, last 6 months, last
year, last 3 years, longer ago)
Motivate ratings for two movies (one seen recently
and one seen more than a year ago)
12. Results
247 users, 4212 ratings
Rating distributions:
Most movies are seen
long ago…
Only 28% seen in the
last year
# Positive ratings
decrease with time
1st timeslot: 60% 4/5
star
Last timeslot: only 36%
*****
****
***
**
*
13. Modeling the ratings
Coefficient Std. Err. t-value
intercept 2.95 0.15 19.05
time 0.29 0.13 2.31
highrated 1.62 0.22 7.43
time2 -0.09 0.02 -3.55
Time x highrated -0.73 0.18 -4.10
Tine2 x highrated 0.11 0.03 3.26
Multilevel model:
Random intercepts for movies
and users
high-rated versus low-rated
shows a different pattern
Regression towards the mean?
High-rated
Low rated
14. This is a problem…
How can we train a recommender system..
If ratings depend on our memory this much…
This is new to psychology as well!
Memory effects like this have not been studied…
Problem lies partly in the type of judgment asked:
Rating is separate evaluation on an absolute scale…
Lacks a good reference/comparison
Two solutions we explored:
Rating support
Different elicitation methods: choice!
15. Joint versus Separate Evaluation
Evaluations of two job candidates for a computer
programmer position expecting the use of a special
language called KY.
Mean WTP (in thousands):
Separate $ 32.7 k $ 26.8 k
Joint $ 31.2 k $ 33.2 k
Candidate A Candidate B
Education B.Sc. computer Sc. B.Sc. computer Sc.
GPA (0-5) 4.8 3.1
KY Experience 10 KY programs 70 KY programs
17
16. Rating support interfaces
Using movielens!
Can we help users during rating to make
their ratings more stable/accurate?
We can support their memory for the movie using tags
We can help ratings on the scale with previous ratings
Movielens has a tag genome and a history of
ratings so we can give real-time user-specific
feedback!
Nguyen, T. T., Kluver, D., Wang, T.-Y., Hui, P.-M., Ekstrand, M. D., Willemsen, M.
C., & Riedl, J. (2013). Rating Support Interfaces to Improve User Experience
and Recommender Accuracy. RecSys 2013 (pp. 149–156)
17. Tag Interface
Provide 10 tags
that are relevant
for that user
and that describe
the movie well
Didn’t really help…
18. Exemplar Interface
Support rating on the
scale by providing
exemplars:
Exemplar: Similar movies
rated before by that user
for that level on the scale
This helps to anchor the
values on the scale better:
more consistent ratings
19. Bur what are preferences?
Ratings are absolute statements
of preference…
But preference is a relative
statement…
I like Grand Budapest hotel more
than King’s Speech
So why not ask users to choose?
Which do you prefer?
20. Others also tried different PE methods
Loepp, Hussein & Ziegler
(CHI 2014)
Choose between sets of
movies that differ a lot on a
latent feature
Chang, Harper & Terveen
(CSCW 2015)
Choose between groups
of similar movies
By assigning points per
group (ranking!)
21. Choice-based preference elicitation
Choices are relative statements that are easier to make
Better fit with final goal: finding a good item rather than making a
good prediction
In Marketing, conjoint-based analysis uses the same idea
to determine attribute weights and utilities based on a
series of (adaptive) choices
Can we use a set of choices in the matrix factorization
space to determine a user vector in a stepwise fashion?
Graus, M.P. & Willemsen, M.C. (2015). Improving the user
experience during cold start through choice-based preference
elicitation. In Proceedings of the 9th ACM conference on
Recommender systems (pp. 273-276)
22. Dimensions in Matrix Factorization
Dimensionality reduction
Users and items are
represented as vectors on a set
of latent features
item vector: utility of attributes
user vector: weights of attributes
Rating is the dot product of
these vectors (overall utility!)
Gus will like Dumb and
Dumber but hate Color Purple
Koren, Y., Bell, R., and Volinsky, C. 2009. Matrix Factorization
Techniques for Recommender Systems. IEEE Computer 42, 8, 30–37.
23. How does this work? Step 1
Latent Feature 1
LatentFeature2
Iteration 1a: Diversified choice set is
calculated from a matrix factorization
model (red items)
Iteration 1b: User vector (blue arrow) is
moved towards chosen item (green item),
items with lowest predicted rating are
discarded (greyed out items)
24. How does this work? Step 2
Iteration 2: New diversified choice set
(blue items)
End of Iteration 2: with updated vector
and more items discarded based on
second choice (green item)
25. Evaluation of Preference Elicitation
Choice-based PE: choosing 10 times from 10 items
Rating-based PE: rating 15 items
After each PE method they evaluated the interface on
interaction usability in terms of ease of use
e.g., “It was easy to let the system know my preferences”
Effort: e.g., “Using the interface was effortful.”
effort and usability are highly related (r=0.62)
Results: less perceived effort for choice-based PE
perceived effort goes down with completion time
26. Behavioral data of PE-tasks
Choice-based PE: most users find their perfect item
around the 8th / 9th item and they inspect quite some
unique items along the way
Rating-based: user inspect many
lists (Median = 13), suggesting
high effort in rating task.
27. Perception of Recommendation List
Participants evaluated each recommendation list
separately on Choice Difficulty and Satisfaction
Satisfaction
with Chosen
Item
Obscurity
Difficulty
Intra List
Similarity
-2.407(.381)
p<.001
-.240 (.145)
p<.1
-.479 (.111)
p<.001
-.257 (.045)
p<.001
14.00 (4.51)
p<.01
Choice-
Based List
+
-
- -
-
28. Conclusion
Participants experienced reduced effort and increased
satisfaction for choice-based PE over rating-based PE
relative (choice) rather than absolute (rating) PE could alleviate the
cold-start problem for new users
Further research needed:
the parameterization of the choice task
strong effect of choice on the popularity of the resulting list
Using trailers helps to decrease popularity (-> IntRS 2016)
novelty effects might have played a role: fun way of interacting?
30. Behavioral change
Behavioral change is hard…
Exercising more, eat healthy, reduce alchohol
consumption (reducing Binge watching on Netflix )
Needs awareness, motivation and commitment
Combi model:
Klein, Mogles, Wissen
Journal of Biomedical Informatics, 2014
31. What can recommenders do?
Persuasive Technology: focused on how to help
people change their behavior:
personalize the message…
Recommenders systems can help with what to
change and when to act
personalize what to do next…
This requires different models/algorithms
our past behavior/liking is not what we want to do now!
Two illustrations of new approach:
energy saving
hypertension management
33. Our first (old) recommender system
Recommendations
Selected
measures
Things you
already do or
don’t want to
Attributes
Set
attribute
weights
Show items with highest Uitem,user, where
Uitem,user = ∑ Vitem,attribute • Wattribute,user
34. Study 3 (AMCIS 2014)
Online lab study
—147 paid pps (79M, 68F, mean age: 40.0)
—Selected pps interacted for at least 2.5 minutes
3 PE-methods, 2 baselines
—Attribute-based PE
—Implicit PE
—Hybrid PE (attribute + implicit)
—Sort (baseline, not personalized)
—Top-N (baseline, not personalized)
http://bit.ly/amcis14
35. Study 3 — Results
Experts prefer Attribute-based PE and Hybrid PE,
novices prefer Top-N and Sort (baselines)
System satisfaction mediates the effect on choice
satisfaction and behavior!
Systemsatisfaction
Domain knowledge http://bit.ly/amcis14
37. Towards a better (psychometric) user model
consumers differ in energy-saving capabilities, attitudes,
goals, …
Our prior work did not take that into account…
Energy-saving interventions are more effective when
personalized. But how?
≠
(Cf. Abrahamse et al., 2005)
38. Single energy saving
dimension/attitude?
Campbell’s Paradigm (Kaiser et al., 2010)
“One’s attitude or ability becomes apparent
through its behavior…”
“Attitude and Behavior are two sides of the
same coin…”
Different from standard psychological
approaches to measure attitudes & intentions
with likert scales…
41
39. Psychological assumptions
Three assumptions for our user model
(Based on Kaiser et al., 2010)
1. All Energy-saving behaviors form a class serving a
single goal: Saving Energy
2. Less performed behaviors yield higher Behavioral
Costs (i.e. are more difficult)
3. Individuals that execute more energy-saving behaviors
have a higher Energy-saving Ability (i.e. more skilled)
40. The Rasch model
The Rasch model equates behavioral difficulties and
individual propensities in a probabilistic model
Log-odds of engagement levels (yes/no):
𝜽 = an individual’s propensity/attitude
𝜹 = behavioral difficulty
P = probability of individual n engaging in behavior i
Rasch also determines individual propensities and
item difficulties & fits them onto a single scale
One scale may have lots of different difficulty levels
43
𝐥𝐧
𝑷 𝒏𝒊
𝟏 − 𝑷 𝒏𝒊
= 𝜽 𝒏 − 𝜹𝒊
41. The resulting Rasch scale
Pictures disclaimer: courtesy of my PhD student
(the guy on the right)
42. Item difficulty vs. person ability
Probability of a person executing behavior
depends on the Ability - Costs
LED lighting
Unplug chargers
Install PV panels
43. Using Rasch for tailored advice
Earlier research (Kaiser, Urban) found evidence
for a unidimensional scale, but with few items &
no advice
We set out a Rasch-based, energy
recommender system that:
Shows the measures in order of difficulty (either
ascending or descending)
Provide tailored conservation advice to users (or not)
Include a more extensive set of measures
46
44. Energy saving ‘Webshop’
Ordered set of measures with rich information and
(in some conditions) recommendations
Order:
Ascending or
descending in
Rasch difficulty
Rasch
recommendation:
start at their ability
(3 items highlighted)
or just at the bottom
(no highlights)
45. Procedure
1. Determining Ability: Each
user indicated for 13
measures whether
he/she executed them
2. Show webshop in one of 4 conditions (total N=224)
and ask to select a set of measures (to execute in
next 4 weeks)
3. Survey to measure user experience
4. After 4 weeks: report back on which measures they
implemented (to some extent) N=86
46. Results: Experience (SEM model)
Perceived
Effort
Perceived
Support
Choice
Satisfaction
-0.40*** 0.59***
-0.56**
* p < .05, ** p < .01, *** p < .001.
0.74**
-0.65*
Rasch
recomm.
Low
Ability
Ascending
Order
48. What did they choose?
Rank-ordered logit with
ability – difficulty/costs
difference as predictor
Without recommendations
they select easy
measures (below their
ability)
With recommendations
they choose measures
around their ability
49. Measure Follow-up after 4 weeks
Follow-up depends on relative costs (they are
more likely to do the easier measures…)
51. Hypertension
Hypertension occurs in 30% of the population
Hypertension is without symptoms.
Hypertension is a leading cause of death.
Hypertension can be prevented or treated with
lifestyle change, such as:
Salt intake reduction
Physical activity
Weight control
Alcohol moderation
53. Study 1: Model construction
Online survey
300 participants between 40 and 60 years old
50% hypertensive
Self-reported engagement in 63 health behaviors
about diet, sodium intake and physical activity
Results:
reliable Rasch scale (for both persons and items)
No strong differences between subgroups
Unidimensional: different categories
mix nicely across the scale
55. Study 2: Coaching strategies
The engagement maximization strategy selects
the easiest behaviors (that she does not do yet)
The motivation maximization strategy selects
behaviors with difficulties that match the
individual’s ability (that she does no do yet)
The random control strategy selects behaviors
(not done yet) at random without regard for
difficulty
56. Study 2: design
150 hypertensive users invited online
Questionnaire used to measure user ability and
find behaviors that user should be coached on.
Pairwise comparison between the three
intervention strategies through virtual coaches
58. Conclusions
Engagement maximization (easy behaviors not
done yet) outperforms random most of the time
The rasch order helps!
Knowing the ability helps to elicit better
recommendations by tailoring: for medium ability
individuals
But for other groups it does not matter much…