2021/09/23 by Diego Sánchez-Moreno, Álvaro Lozano Murciego, Sánchez-Moreno, Diego +7
Computer Science · Neuroscience · #Music and Audio Processing #Neuroscience and Music Perception #Speech Recognition and Synthesis
paper · pdf · doi:10.48550/arxiv.2109.11231
Music listening preferences at a given time depend on a wide range of\ncontextual factors, such as user emotional state, location and activity at\nlistening time, the day of the week, the time of the day, etc. It is therefore\nof great importance to take them into account when recommending music. However,\nit is very difficult to develop context-aware recommender systems that consider\nthese factors, both because of the difficulty of detecting some of them, such\nas emotional state, and because of the drawbacks derived from the inclusion of\nmany factors, such as sparsity problems in contextual pre-filtering. This work\ninvolves the proposal of a method for the detection of the user contextual\nstate when listening to music based on the social tags of music items. The\nintrinsic characteristics of social tagging that allow for the description of\nitems in multiple dimensions can be exploited to capture many contextual\ndimensions in the user listening sessions. The embeddings of the tags of the\nfirst items played in each session are used to represent the context of that\nsession. Recommendations are then generated based on both user preferences and\nthe similarity of the items computed from tag embeddings. Social tags have been\nused extensively in many recommender systems, however, to our knowledge, they\nhave been hardly used to dynamically infer contextual states.\n