=Paper=
{{Paper
|id=Vol-1911/14
|storemode=property
|title=Estimate Features Relevance for Groups of Users
|pdfUrl=https://ceur-ws.org/Vol-1911/14.pdf
|volume=Vol-1911
|authors=Stefano Cereda,Leonardo Cella,Paolo Cremonesi
|dblpUrl=https://dblp.org/rec/conf/iir/CeredaCC17
}}
==Estimate Features Relevance for Groups of Users==
https://ceur-ws.org/Vol-1911/14.pdf
Estimate features relevance for groups of users
Stefano Cereda1 , Leonardo Cella1 , and Paolo Cremonesi1
Politecnico di Milano
Abstract. In item cold-start, collaborative filtering techniques cannot
be used directly since newly added items have no interactions with users.
Hence, content-based filtering is usually the only viable option left.
In this paper we propose a feature-based machine learning model that
addresses the item cold-start problem by jointly exploiting item content
features, past user preferences and interactions of similar users. The pro-
posed solution learns a relevance of each content feature referring to a
community of similar users. In our experiments, the proposed approach
outperforms classical content-based filtering on an enriched version of
the Netflix dataset.
1 Introduction and Related Works
Traditional Content Based recommender systems (CBF) need to represent users
and items profiles in order to recommend similar items to those previously liked
by users. Their main advantage is the capability of recommending previously
unseen items, thus they solve the item-cold start issue. On the contrary, Col-
laborative Filtering (CF) algorithms usually reach better performance on pre-
dictions, especially with many interactions between users and items [1]. Their
downside consists of the inability of recommending items with no previous inter-
actions. Even if CBF approaches are able to solve the item cold-start problem,
they are affected by at least two relevant limitations: recommended items tend
to be too similar to previously rated items (over-specialization problem) and
recommendations do not depend on preferences of similar users.
Some attempts to improve recommendation quality of CBFs consists of: Fil-
tering methods and Embedded approaches. The former main drawback is that
they do not take into account the ratings of users, therefore ignoring if the
feature-based similarity between items is aligned with the user perception of
similarity ([2], [3]). Embedded approaches perform feature weighting during the
learning process and use its objective function to guide searching for relevant
features. Instances of this methodology are: SSLIM [5] , UFSM [4] and Fac-
torization Machines. The main drawback of embedded methods is the coupling
between the collaborative and content components of the model. When used on
datasets with unstructured user-generated features (e.g., tags) the noise from
the features propagate to the collaborative part, affecting the overall prediction
quality.
As a first solution to this problem, we have developed a machine learning
algorithm whose aim is to compute global1 feature weights based on a pure item
1
in the sense that the relevance scores were shared by all the different users.
collaborative filtering approach. Its main objective was to embed in item features
also information regarding user interests. In this research we propose an exten-
sion to this approach, the main contribution brought by this work is a general,
straightforward wrapper to make content-based methods rate-aware and based
on communities of similar users. Our experiments are conducted on the Netflix
dataset in a version enriched with IMDB attributes. The experiments shown
that the proposed solution outperforms classical pure content-based approaches.
2 Clustered Feature Weighting
Our objective is to recommend items from a set I to users in a set U . Items
are described by the d-dimensional set of features F. User interactions are col-
lected with the R|U |×|I| feedback matrix. Item features are described by the
item features matrix A|I|×|F | , aij = 1 iff item i has feature j. In general, user-
cluster based recommender systems rely on a cluster-dependent similarity matrix
|I|×|I|
Sp , where p denotes the considered subset of users.
The predicted rate of user u, that belongs to the group pu , for item i is computed
as follows: P pu
j∈Nkpu (i) ruj sij
r̂ui = P pu (1)
j∈N pu (i) sijk
where: spijuis a local item-item similarity derived from the user subset pu to
which the target user u belongs and Nkpu (i) is the set of k nearest neighbors
of item i according to the similarity model of cluster pu . Starting from this
model, we would recommend the items whose predicted ratings are the largest.
Feature weighting aims to derive a feature vector wpu ∈ R|F | such that each
entry wlpu ∈ wpu reflects the lth feature relevance for the pu subset of users. We
define the weighted similarity spiju between items i and j for the cluster pu as:
X
spiju = wfpu aif ajf = hwpu , ai aj i (2)
f ∈F
|F |
where ai , aj ∈ {0, 1} are the feature vectors of items i and j respectively and
is the element-wise product. We propose to compute the feature weights by
solving the following LSQ problem for each cluster of users pu :
X X
argmin ||scCF
ij − spiju ||2 (3)
w∗pu
i∈I j∈I\{i}
More specifically, in our experiments we have adopted LSLIMr0 [6] as local
similarity matrix S cCF and CLUTO [7]2 to derive the user subsets pu .
Since our goal is to learn a set of feature weights so that CBF similarities mimic
as close as possible CF ones, there is no need to add a regularization term,
2
this choice is based on the methodology followed in [6].
thus greatly simplifying the optimization. Experimental results confirmed this
hypothesis.
When a new item is added to the catalog, we use w∗ pu to compute its
weighted similarity w.r.t. the previously existing items. Then, it can be rec-
ommended to users belonging to subset pu by using Equation 1. We call the
proposed approach CLFW (Clustered Least-square Features Weighting).
3 Experimental Evaluation
Dataset. For our experiments, we used a version of the Net- Fig. 1. Dataset
flix dataset enriched with structured and unstructured at- partitioning.
tributes extracted from IMDB. This dataset has 186K users, Items
6.5k movies and 6.7M ratings in 1-5 scale. The rating data 4866 1623
is enriched with 16803 binary attributes representing various A
B
kinds of meta-information on movies such as director, actor, A1 A2
genres and user-generated tags3 . To investigate the new-item
scenario, we performed a 70/30 random hold-out split over
items as shown in Figure 1. The sub matrix A has then been divided by moving
30% of positive (> 3) ratings into A2 and everything else in A1. A1 has then
been used to compute LSLIMr0 and therefore to fit CLFW. When evaluating
the warm-start scenario we used A1 as user profiles and A2 as ground truth,
whereas for the cold-item we used the positive ratings of B as ground truth and
A1 as user profiles.
Baselines. As in the previous work we have used simple unweighted cosine sim-
ilarity (Cos) and TF-IDF-weighted cosine similarity (CosIDF) as CBF baselines
to evaluate the performance of CLSFW in both scenarios.
Performance Analysis In Table 1, we report the RMSE computed over predicted
rates for different neighborhood sizes k in the new-item scenario. The warm-start
scenario is instead represented by Table 2.
Table 1. RMSE evaluation for the new- Table 2. RMSE evaluation for the warm-
item scenario. start scenario.
k CLFW regCLFW Cos CosIDF k LSLIMr0 CLFW regCLFW Cos CosIDF
50 1.321 1.714 2.317 2.612 50 1.372 1.838 1.962 2.634 2.901
100 1.426 1.409 1.923 2.160 100 1.208 1.522 1.619 2.209 2.477
200 1.188 1.182 1.538 1.756 200 1.129 1.291 1.362 1.821 2.058
We can state that in both scenarios CLFW consistently outperforms both
the baselines on RMSE at any value of k. Moreover, in the warm-start scenario,
3
the set of content features was significantly augmented with respect to our previous
unclustered work.
it is nearly as good as LSLIMr0. We want to also highlight that CLFW differs
from the other CBF baselines solely in the feature weighting scheme. Therefore,
the improvement in performance must be due to a better feature weighting dis-
covered by our approach. By comparing the CLFW column with the regCLFW
one 4 , we can observe that the regularization does not bring a performance im-
provement. This is reasonable and totally in agreement with our prediction. In
fact, the data from which we are learning do not contain noise and, further more,
the number of weights that we learn does not allow to overestimate the model
complexity.
4 Conclusions and Future Work
With this research we investigated the possibility of deriving a user based fea-
ture weighting. We have presented ongoing results of an extended approach that
solves the item cold-start issue by defining personalized features relevance. The
ongoing development is focused in the usage of different personalization method-
ologies and extension to other datasets. Moreover, we are interested in combining
this clustered approach with the, already developed, global one.
References
[1] Istvn Pilszy, Domonkos Tikk: Recommending new movies: even a few ratings
are more valuable than metadata RecSys 09 Proceedings of the third ACM
conference on Recommender systems , 93–100 (2009)
[2] Lops Pasquale , De Gemmis Marco , Semeraro Giovanni: Content-based recom-
mender systems: State of the art and trends Recommender systems handbook
, 73–105 (2011)
[3] Panagiotis Symeonidis, Alexandros Nanopoulos, Yannis Manolopoulos:
Feature-Weighted User Model for Recommender Systems UM 07 Proceedings
of the 11th international conference on User Modeling , Springer, 10.1007/978-
3-540-73078-1-13 97 – 106 (2007)
[4] Elbadrawy Asmaa, Karypis, George: User-Specific Feature-Based Similarity
Models for Top-n Recommendation of New Items, ACM Trans. Intell. Syst.
Technol., 6, 33:1–33:20, (May 2015) 10.1145/2700495, ACM
[5] Ning Xia , Karypis George: Sparse Linear Methods with Side Information for
Top-N Recommendations, Proceedings of the 21st International Conference on
World Wide Web , WWW ’12 Companion, ACM 581 – 582, (2012)
[6] Christakopoulou Evangelia, Karypis George: Local Item-Item Models for Top-
N Recommendation, 10th ACM Conference on Recommender Systems , Rec-
Sys16 67 – 74 , (2016)
[7] http://glaros.dtc.umn.edu/gkhome/cluto/cluto/overview
4
which contains the results of our algorithm when the feature weights are computed
adding an l2 regularization term to Equation 3