An Annotated Corpus of Dota 2 In-Game Chat, Labeled for Emotional States of Anxiety and Boredom

An Annotated Corpus of Dota 2 In-Game Chat, Labeled for Emotional States of Anxiety and Boredom Abstract This dataset provides a valuable corpus of 8,000 in-game chat messages from the popular Multiplayer Online Battle Arena (MOBA) game, Dota 2. The messages have been manually annotated to classify the emotional state of the player, specifically focusing on expressions of anxiety and boredom. Given the high-stress, competitive, and team-dependent nature of Dota 2, player communication is a rich source of authentic, spontaneous emotional expression. This dataset is designed to support research in natural language processing (NLP), computational linguistics, psychology, and human-computer interaction (HCI) for tasks related to emotion detection, sentiment analysis, and the study of player behavior in digital environments. Description of Data The data consists of textual messages exchanged between players during live Dota 2 matches. Dota 2 is a complex, real-time strategy game where two teams of five players compete to destroy the opposing team's base. Matches are often long (30-60 minutes on average) and require intense teamwork, strategic thinking, and mechanical skill. This high-stakes environment frequently elicits strong emotional responses from players, which are often articulated in the in-game chat. The language captured in this dataset is informal and characteristic of online gaming communities, including the use of game-specific jargon, slang, abbreviations, typos, and emotionally charged statements. The corpus serves as a real-world example of communication under pressure. Methodology The dataset was collected from publicly available Dota 2 match data. Following collection, the corpus was manually annotated by trained annotators to categorize each message based on the dominant emotional theme it conveyed. Each message was assigned one of two primary labels: Anxiety: This label was applied to messages reflecting frustration, anger, stress, helplessness, blaming teammates, or a sense of being overwhelmed. These messages often pertain to negative game events, perceived mistakes (one's own or others'), or the pressure of a difficult match. Examples: \"no chance to win with braindead afk lc\", \"I HATE MY TEAM SO BAD\", \"my hands are shaking\" Boredom: This label was applied to messages indicating disengagement, a lack of challenge, or the feeling that the game is pointless or already decided. These messages often include taunts about the opponent's lack of skill or a desire for the match to end quickly due to its uncompetitive nature. Examples: \"no wonder it was so easy\", \"this game is fucking easy\", \"just finish this crap\" The annotation process focused on the contextual meaning of the message within a competitive gaming framework. File Description The data is provided in a single CSV (Comma-Separated Values) file named dota2_chat_emotion_dataset.csv. The file contains the following columns: chat_message: The raw text of the chat message. class: The annotated label for the message ('anxiety' (1) or 'boredom' (0)). Potential Use Cases This dataset is suitable for a wide range of research purposes, including but not limited to: Natural Language Processing: Training, testing, and validating models for emotion detection, sentiment analysis, and toxicity classification in niche, high-jargon domains. Psychology & Cognitive Science: Studying emotional regulation, stress responses, and group dynamics in online and competitive settings. Human-Computer Interaction (HCI) & Game Studies: Investigating player experience (PX), designing systems to mitigate negative player interactions, and understanding communication patterns within online communities. Sociolinguistics: Analyzing the use of slang, jargon, and community-specific language under different emotional conditions.