Decoding Affect in Dyadic Conversations: Leveraging Semantic Similarity through Sentence Embedding