Capturing the Moment: Lightweight Similarity Computations

Abstract

Similarity computations are crucial in various web activities like advertisements, search or trust-distrust predictions. These similarities often vary with time as product perception and popularity constantly change with users' evolving inclination. The huge volume of user-generated data typically results in heavyweight computations for even a single similarity update. We present I-SIM, a novel similarity metric that enables lightweight similarity computations in an incremental and temporal manner. Incrementality enables updates with low latency whereas temporality captures users' evolving inclination. The main idea behind I-SIM is to disintegrate the similarity metric into mutually independent time-aware factors which can be updated incrementally. We illustrate the efficacy of I-SIM through a novel recommender (SWIFT) as well as through a trust-distrust predictor in Online Social Networks (I-TRUST). We experimentally show that I-SIM enables fast and accurate predictions in an energy-efficient manner.

The Imitation Game: Algorithm Selection by Exploiting Black-Box Recommenders

“…The overall precision over the whole test set is the average over the precision values for all users in the test set. Note that a recommended item is considered as a hit, if the user rates that item anytime later than the time of the recommendation with a rating score larger than 50% of the maximum score [5].…”

“…A question that may arise in our problem setup is whether there are any constraints for the relation between the training data of the candidate algorithms (local) and the training data of the black-box. In our evaluation ( §3), this relation is that the data has no overlap but comes from the same dataset, i.e., the user rating behaviour follows the preference and behavioural drifts of the MovieLens dataset [5]. We plan to evaluate the performance of RECRANK under alternative relation scenarios (e.g., the data comes from different datasets) as part of our future work.…”

“…-Edge e ij : Item j is at least in one of the top-N recommendation lists given as an output to a query for item i. where eij ranking score is the summation of the recommender output over all recommendations for item j triggered by a query for item i. 5 The edge weight captures the fact that there are typically multiple recommendations between the same items. For example, a user might receive the same item in multiple top-N recommendations before she either clicks it or the recommender lowers its ranking and removes it from the top-N list.…”

BoostJet: Towards Combining Statistical Aggregates with Neural Embeddings for Recommendations

“…Likewise, individuals gradually change their taste in movies and music. There have been also several work on incorporating such temporality in recommendations [16,6,12,9]. Demography.…”

“…More precisely, the nodes at the same depth in M N share not only a feature but also a border. Typically, a decision tree in M N has 6 di erent splits and 2 6 leaves. Such trees can be interpreted as a p-dimensional matrix indexed by a p-dimensional boolean variable, where p is the number of di erent splits in the tree.…”