A User Training Error based Correction Approach combined with the Synthetic Coordinate Recommender System

Costas Panagiotakis1,3, Harris Papadakis2 and Paraskevi Fragopoulou2,3

1Department of Management Science and Technology, Hellenic Mediterranean University, Greece
2Department of Electrical and Computer Engineering, Hellenic Mediterranean University, Greece
3Institute of Computer Science, FORTH, Greece

  Paper: https://doi.org/10.1145/3386392.3397591

Video Pres.:  https://www.youtube.com/watch?v=dcPRCdOK6pM

We propose a Synthetic Coordinate Recommendation system using a user Training Error based Correction approach (SCoR-UTEC). Synthetic Euclidean coordinates are assigned by SCoR system to users and items, so that, when the system converges, the distance between a user and an item provides an accurate prediction of the user’s preference for that item. After the SCoR execution, we introduce a stage called UTEC to correct the SCoR recommendations. The experimental results demonstrate the high performance of the proposed second stage on real world datasets.

 

 

  • The proposed parameter free, second stage is executed after the SCoR method to improve the SCoR based recommendations, called User Training Error based Correction approach (UTEC).
  • The recommendations of SCoR are improved by the UTEC taking into account the error on the training set between users and items and their proximity in the synthetic Euclidean space of SCoR.
  • UTEC is also applicable on any model-based recommender system with positive training error like SCoR.

SCoR

  • SCoR [1] algorithm is a Synthetic Euclidean Coordinates system, which randomly assigns a position in an N-dimensional Euclideanspace to each element in the user and the item sets.
  • The distance between user u and item i corresponds to the prediction for the preference of user u for item i. The item preferences for the user located in the center of the graph is indicated by the brightness of the graph background – from white (like) to black (dislike).
  • It has been successfully applied to movie recommendation [1], personalized video summarization [2], detection of abnormal profiles in RS [3], community detection [4], and to the interactive image segmentation problem [5].

UTEC

The input of is the training set of items for user u that are given in the model based RS e.g. SCoR, TRu = {i1, …, i |TRu| }.

  • ru(ik ) and ru‘(ik) denote the rating of user u for item ik according to the ground truth and the recommendation of the SCoR system, respectively.
  • p(.) denotes the coordinates in the synthetic Euclidean space provided by SCoR.
  • The key idea of the UTEC is that the final error (after UTEC execution) on any sample of the training set ik ∈ TRu should be zero.
    • It holds that, cu (ik) = eu(ik), ∀ ik ∈ TRu, where cu(i) is the recommendation correction of the UTEC approach and eu (ik) denotes the training recommendation error of user u and item ik ∈ TRu.

 

 

 

A synthetic example after the execution of SCoR that shows the positions of nodes in R2, the corrections from a training sample and the corrections provided by UTEC.

 

EXPERIMENTAL RESULTS

 

REFERENCES

[1] Harris Papadakis, Costas Panagiotakis, and Paraskevi Fragopoulou. 2017. SCoR: A Synthetic Coordinate based System for Recommendations. Expert Systems with Applications 79 (2017), 8–19.

[2] Costas Panagiotakis, Harris Papadakis, and Paraskevi Fragopoulou. 2020. Personalized Video Summarization based exclusively on User Preferences. In European Conference on Information Retrieval.

[3] Costas Panagiotakis, Harris Papadakis, and Paraskevi Fragopoulou. 2020. Unsupervised and Supervised Methods for the Detection of Hurriedly Created Profiles in Recommender Systems. Machine Learning and Cybernetics (2020).

[4] Harris Papadakis, Costas Panagiotakis, and Paraskevi Fragopoulou. 2014. Distributed detection of communities in complex networks using synthetic coordinates. Journal of Statistical Mechanics: Theory and Experiment 2014, 3 (2014), P03013.

[5] Costas Panagiotakis, Harris Papadakis, Elias Grinias, Nikos Komodakis, Paraskevi Fragopoulou, and Georgios Tziritas. 2013. Interactive image segmentation based on synthetic graph coordinates.  Pattern Recognition 46, 11 (2013), 2940–2952.