In the last decade, driven also by the availability of an unprecedented computational power and storage capabilities in cloud environments, deep learning emerged as one of the most promising approaches in the generation and training of models that can be applied to a wide variety of application elds. In this paper, we instigate how to exploit the semantic information encoded in a knowledge graph to build connections between units in a Neural Network, thus leading to a semantics-aware autoencoder, SEM-AUTO, able to extract and weight semantic features that can eventually be used to build a recommender system. We tested how our approach behaves in the presence of cold users on the MovieLens 1M dataset and compare results with BPRSLIM.
Recommender systems (RS) are essential tools that help users to nd those items that are relevant to them. Collaborative ltering (CF) approaches are able to provide very accurate recommendations, especially if many data about usersitems interactions are available, but they fail when users rate a few items (cold users) or items have a few ratings (cold items).
Among the various deep learning approaches, autoencoders are a particular
con guration of neural networks which have recently attracted attention in the
RS community. In [
A novel idea to tackle the recommendation problem was introduced in [
IIR 2018, May 28-30, 2018, Rome, Italy. Copyright held by the author(s).
In this paper we present how to to build a user pro le by leveraging both autoencoders and semantic information available a knowledge graph, whose structure is further exploited to draw the topology of the underlying neural network. Each neuron in the hidden layer represents a class or a category associated to an item; the resulting neural network is trained to autoencode the ratings of each user. Therefore, every hidden neuron weight is interpreted as the relevance the corresponding feature has for that user. Eventually, the vectors of feature weights are used to compute a top-N recommendation list. 2
Arti cial Neural Networks (ANNs) are computational models originally proposed to catch underlying relationships in a set of data by using positive and negative examples fed into the network (supervised learning). Roughly, in an autoencoder network one tries to \predict" x from x. The idea is to rst compress (encode) the input vector to t in a smaller feature space, and then try to reconstruct (decode) it back. This means that the model learns in the hidden layers, a representation of the input and therefore a latent representation of the knowledge behind the input data.
Semantics-aware autoencoders Autoencoders, just like other methods for latent representation, are unable to provide an explanation for the latent factors they provide. To address this issue, we propose to give a meaning to connection with the hidden layer and to its neurons by exploiting semantic information explicitly available in knowledge graphs. The main idea of the SEM-AUTO approach is to map connections between units from layer i to layer i+1, re ecting the nodes available in a knowledge-based graph (KG). In particular, we mapped the autoencoder network topology with the categorical information related to items rated by users. As the nodes in the hidden layer correspond to categories in the knowledge graph, once the model has been trained, the sum of the weights of edges entering a node represents somehow its worthiness in the de nition of a rating. If we consider the nodes associated (connected) to a speci c item, their weight may be considered as an initial form of explanation for a given rating. Please note, that such autoencoders do not need bias nodes because these latter are not representative of any semantic data in the graph.
Once the network converges we have a latent representation of features associated to a user pro le together with their weight. However, very interestingly, this time the features also have an explicit meaning as they are in a one to one mapping with elements (nodes) in a knowledge graph. Our autoencoder is therefore capable of learning for each user the semantics behind her ratings and eventually weight them. Given the trained autoencoder, a user pro le is then built by considering the features associated to items she rated in the past. Recommendation. As we said before, the weight associated to a feature fn is the summation of the weights wjn computed in the semantic autoencoder for each edge entering the node representing the feature itself. As we train an auto Title Suppressed Due to Excessive Length encoder for each user, we have weights changing depending on the original user pro le P (u) = fhi; rig with i being an item rated by the user and r its associated rating. More formally, we have w(fn; u) = j=inndeg(fn)
X j=0 wjn where inndeg(fn) is the number of edges entering the node representing the feature fn. Due to the high sparseness of the feature-item matrix, we exploited collaborative information available in the original dataset to further enhance user pro les. We projected them in a Vector Space Model where each feature is a dimension of the vector space and computed the cosine similarity between users and, for each user we computed the set K(u) containing the k users most similar to u. Hence, we can infer the value of missing features for user u with its average from her neighborhood. After this post-processing step, ratings for unknown items ~i to u can be computed by just sum item's features with the weights from user pro le u. 3
In this section, we present the experimental evaluations on MovieLens 1M dataset, focused on cold-users with a number of ratings equal to 2 or 5.
#ratings k f1@10 precision@10 recall@10 nDCG@10 ERR-IA@10 SBEPMRS-ALUIMTO 2 10 00.0.02231074916623823 00.0.03332064496030578 00.0.01176729571049297 00.0.02283235833537768 00..00128680205733941 SBEPMRS-ALUIMTO 5 100 00..003398057988955345 00.0.05540006369272355 00..003320004260255321 00.0.04485615283692493 00.0.04472112214376197 Table 1: Experimental Results. #ratings represents the number of ratings in cold users. k is the number of similar users belonging to K(u).
In our experiments, we compared our approach with the implementation of
BPRSLIM [
http://mymedialite.net
In this paper, we have presented a novel method to design semantics-aware autoencoders (SEM-AUTO) driven by information encoded in knowledge graphs. We compute a latent representation of items and attach an explicit semantics to selected features. This allows our system to exploit the power of deep learning techniques and to have a meaningful and understandable representation of the trained model. We used our approach to autoencode user ratings in a recommendation scenario via the DBpedia knowledge graph and proposed a simple algorithm to compute recommendations based on the semantic features we extract with our autoencoder. Experimental results show that even with a simple approach that just sums the weights associated to features we are able to beat state of the art recommendation algorithms for cold user scenarios.