Michael D. Porter

Moeen Mostafavi, Michael D. Porter

This paper focuses on the EmoWoz dataset, an extension of MultiWOZ that provides emotion labels for the dialogues. MultiWOZ was partitioned initially for another purpose, resulting in a distributional shift when considering the new purpose of emotion recognition. The emotion tags in EmoWoz are highly imbalanced and unevenly distributed across the partitions, which causes sub-optimal performance and poor comparison of models. We propose a stratified sampling scheme based on emotion tags to address this issue, improve the dataset's distribution, and reduce dataset shift. We also introduce a special technique to handle conversation (sequential) data with many emotional tags. Using our proposed sampling method, models built upon EmoWoz can perform better, making it a more reliable resource for training conversational agents with emotional intelligence. We recommend that future researchers use this new partitioning to ensure consistent and accurate performance evaluations.

Moeen Mostafavi, Michael D. Porter, Dawn T. Robinson

Predicting the outcome of a process requires modeling the system dynamic and observing the states. In the context of social behaviors, sentiments characterize the states of the system. Affect Control Theory (ACT) uses sentiments to manifest potential interaction. ACT is a generative theory of culture and behavior based on a three-dimensional sentiment lexicon. Traditionally, the sentiments are quantified using survey data which is fed into a regression model to explain social behavior. The lexicons used in the survey are limited due to prohibitive cost. This paper uses a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model to develop a replacement for these surveys. This model achieves state-of-the-art accuracy in estimating affective meanings, expanding the affective lexicon, and allowing more behaviors to be explained.

Moeen Mostafavi, Michael D. Porter

During online chats, body-language and vocal characteristics are not part of the communication mechanism making it challenging to facilitate an accurate interpretation of feelings, emotions, and attitudes. The use of emojis to express emotional feeling is an alternative approach in these types of communication. In this project, we focus on modeling a customer's emotion in an online messaging session with a chatbot. We use Affect Control Theory (ACT) to predict emotional change during the interaction. To let the customer use emojis, we also extend the affective dictionaries used by ACT. For this purpose, we mapped Emoji2vec embedding to the affective space. Our framework can find emotional change during messaging and how a customer's reaction is changed accordingly.