AttendAffectNet–Emotion Prediction of Movie Viewers Using Multimodal Fusion with Self-Attention

Abstract

In this paper, we tackle the problem of predicting the affective responses of movie viewers, based on the content of the movies. Current studies on this topic focus on video representation learning and fusion techniques to combine the extracted features for predicting affect. Yet, these typically, while ignoring the correlation between multiple modality inputs, ignore the correlation between temporal inputs (i.e., sequential features). To explore these correlations, a neural network architecture—namely AttendAffectNet (AAN)—uses the self-attention mechanism for predicting the emotions of movie viewers from different input modalities. Particularly, visual, audio, and text features are considered for predicting emotions (and expressed in terms of valence and arousal). We analyze three variants of our proposed AAN: Feature AAN, Temporal AAN, and Mixed AAN. The Feature AAN applies the self-attention mechanism in an innovative way on the features extracted from the different modalities (including video, audio, and movie subtitles) of a whole movie to, thereby, capture the relationships between them. The Temporal AAN takes the time domain of the movies and the sequential dependency of affective responses into account. In the Temporal AAN, self-attention is applied on the concatenated (multimodal) feature vectors representing different subsequent movie segments. In the Mixed AAN, we combine the strong points of the Feature AAN and the Temporal AAN, by applying self-attention first on vectors of features obtained from different modalities in each movie segment and then on the feature representations of all subsequent (temporal) movie segments. We extensively trained and validated our proposed AAN on both the MediaEval 2016 dataset for the Emotional Impact of Movies Task and the extended COGNIMUSE dataset. Our experiments demonstrate that audio features play a more influential role than those extracted from video and movie subtitles when predicting the emotions of movie viewers on these datasets. The models that use all visual, audio, and text features simultaneously as their inputs performed better than those using features extracted from each modality separately. In addition, the Feature AAN outperformed other AAN variants on the above-mentioned datasets, highlighting the importance of taking different features as context to one another when fusing them. The Feature AAN also performed better than the baseline models when predicting the valence dimension.

Construction of university English teaching connection model based on multimedia fusion

“…The main goal of multimodal fusion is to reduce the heterogeneous differences among modalities, [17] keep the integrity of specific semantics of each modality, and achieve the best performance in deep learning models. It is divided into three types: joint architecture, cooperative architecture and codec architecture.…”

Predicting emotion from music videos: exploring the relative contribution of visual and auditory information to affective responses

“…Reviews have summarized extracted features relevant to affect detection in the audio modality such as intensity (loudness, energy), timbre (MFCC) and rhythm (tempo, regularity) features [31] and video modality such as colour, lighting key, motion intensity and shot length [35], [36]. Features that can capture complex latent dimensions in the data, such as the audio embeddings generated by the VGGish model [37] [38] [39], are also becoming increasingly popular. Features may be provided in the dataset or extracted from the source data if it is available.…”

“…Audio feature extraction was performed with openS-MILE [46], a popular open-source library for audio feature extraction. Specifically, we used the "emobase" configuration file to extract a set of 988 low-level descriptors (LLDs) including MFCC, pitch, spectral, zero-crossing rate, loudness and intensity statistics, many of which have been shown to be effective for identifying emotion in music [38], [39], [47], [48]. Many other configurations are available in openSMILE but we provide the "emobase" set of acoustic features since it is well-documented and was designed for emotion recognition applications [49].…”

MERP: A Music Dataset with Emotion Ratings and Raters’ Profile Information

“…Alternatively, in an effort to collect larger quantities of affect labels in a shorter amount of time, although with a potential loss in accuracy, crowd-sourcing on platforms such as Amazon Mechanical Turk (MTurk) has also been explored [ 3 , 53 , 54 , 55 , 56 ]. Some researchers utilize a mix of both online and offline collection methods [ 57 , 58 ], or even use predictive models such as AttendAffectNet [ 59 ] for the emotion labeling [ 60 ]. Regardless of the data collection method, it is important for each musical excerpt in the dataset to be labelled by multiple participants in order to account for subjectivity.…”