Datasets:

havens2

/

naacl2022

like 0

[ "That", "is,", "PM(s)", "cannot", "be", "regarded", "as", "a", "pre-trained", "memory", "because", "the", "parameters", "are", "randomly", "initialized,", "and", "simply", "increasing", "the", "model", "complexity", "does", "not", "help", "to", "improve", "the", "performance." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "CoMPM(s)", "performs", "worse", "than", "CoMPM,", "and", "even", "performs", "worse", "than", "CoM", "on", "the", "other", "datasets", "except", "for", "MELD." ]

[ 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 13 ]

[ "If", "PM", "parameters", "are", "not", "frozen", "and", "are", "instead", "randomly", "initialized", "(i.e.", "PM(s)),", "the", "performance", "deteriorates." ]

[ 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "confirm", "the", "effect", "of", "PM", "structure", "in", "the", "model", "through", "the", "performance", "of", "CoMPM(s)." ]

[ 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1 ]

[ "As", "a", "result,", "we", "show", "that", "the", "combination", "of", "CoM", "and", "PM", "is", "effective", "in", "achieving", "better", "performance." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0 ]

[ "Since", "PM", "does", "not", "consider", "context,", "it", "showed", "worse", "performance", "than", "CoM,", "and", "the", "performance", "gap", "is", "larger", "in", "the", "IEMOCAP", "dataset", "with", "a", "higher", "average", "number", "of", "conversation", "turns." ]

[ 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "effect", "of", "PM", "can", "be", "confirmed", "through", "the", "performance", "comparison", "between", "CoM", "and", "CoMPM,", "and", "the", "effect", "of", "CoM", "can", "be", "confirmed", "by", "comparing", "the", "results", "of", "CoM", "and", "PM." ]

[ 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1 ]

[ "We", "use", "the", "pre-trained", "model", "provided", "by", "the", "site", "4", "as", "PM(k)." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Following", "previous", "studies,", "we", "use", "the", "average", "vector", "for", "each", "token", "in", "PM(k)", "as", "the", "feature", "of", "the", "utterance." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "CoMPM(k)", "is", "a", "model", "in", "which", "PM", "is", "trained", "on", "ConceptNet." ]

[ 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0 ]

[ "CoMPM", "is", "a", "model", "in", "which", "both", "CoM", "and", "PM", "parameters", "are", "updated", "in", "the", "initial", "state", "of", "the", "pre-trained", "LM.", "CoMPM(f)", "is", "a", "model", "in", "which", "PM", "parameters", "are", "frozen", "in", "the", "initial", "state", "(pre-trained", "LM)", "and", "is", "not", "trained", "further,", "and", "CoMPM(s)", "is", "a", "model", "in", "which", "PM", "is", "trained", "from", "scratch." ]

[ 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0 ]

[ "PM", "used", "alone", "predicts", "emotion", "only", "with", "the", "utterance", "of", "the", "current", "turn", "without", "considering", "the", "context." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "PM", "used", "alone", "is", "not", "used", "as", "a", "memory", "module,", "but", "the", "same", "backbone", "is", "used." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "CoM", "used", "alone", "does", "not", "leverage", "PM", "and", "predicts", "emotions", "by", "considering", "only", "the", "dialogue", "context." ]

[ 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Table", "2", "shows", "the", "performance", "of", "the", "previous", "methods", "and", "our", "models." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "4.4", "Result", "and", "Analysis" ]

[ 0, 0, 0, 0 ]

[ "3https://github.com/something678/TodKat" ]

[ 0 ]

[ "The", "performance", "of", "ToDKAT", "in", "MELD", "was", "re-released", "on", "github", "3." ]

[ 0, 0, 0, 1, 0, 13, 0, 0, 0, 0, 0 ]

[ "ToDKAT", "(Zhu", "et", "al.,", "2021)", "proposes", "a", "language", "model", "with", "topic", "detection", "added,", "and", "improves", "performance", "by", "combining", "it", "with", "commonsense", "knowledge." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "DialogueCRN", "(Hu", "et", "al.,", "2021)", "introduces", "an", "intuitive", "retrieving", "process,", "the", "reasoning", "module,", "which", "understands", "both", "situation-level", "and", "speakerlevel", "contexts." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Psychological", "performs", "emotion", "recognition", "by", "utilizing", "intention", "of", "utterances", "from", "not", "only", "past", "contexts", "but", "also", "future", "context." ]

[ 1, 0, 11, 12, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Psychological", "(Li", "et", "al.,", "2021)", "uses", "commonsense", "knowledge", "as", "enrich", "edges", "and", "processes", "it", "with", "graph", "transformer." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "ERMC-DisGCN", "(Sun", "et", "al.,", "2021)", "proposes", "a", "discourse-aware", "graph", "neural", "network", "that", "utilizes", "self-speaker", "dependency", "of", "interlocutors", "as", "a", "relational", "convolution", "and", "informative", "cues", "of", "dependent", "utterances", "as", "a", "gated", "convolution." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "This", "model", "uses", "pre-trained", "RoBERTa", "as", "a", "feature", "extractor", "and", "leverages", "COMET", "trained", "with", "ATOMIC", "as", "the", "commonsense", "knowledge." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "COSMIC", "(Ghosal", "et", "al.,", "2020)", "incorporates", "different", "elements", "of", "commonsense", "such", "as", "mental", "states,", "events", "and", "causal", "relations,", "and", "learns", "the", "relations", "between", "participants", "in", "the", "conversation." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Models" ]

[ 0 ]

[ "HiTrans", "(Li", "et", "al.,", "2020)", "proposes", "a", "transformerbased", "context-", "and", "speaker-sensitive", "model." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "RGAT+P", "(Ishiwatari", "et", "al.,", "2020)", "(relational", "graph", "attention", "networks)", "proposes", "relational", "position", "encodings", "with", "sequential", "information", "reflecting", "the", "relational", "graph", "structure,", "which", "shows", "that", "both", "the", "speaker", "dependency", "and", "the", "sequential", "information", "can", "be", "captured." ]

[ 1, 0, 0, 0, 0, 1, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "(2020)", "shows", "the", "performance", "of", "RoBERTa+DialogueRNN", "when", "the", "vectors", "of", "the", "tokens", "are", "extracted", "with", "a", "pretrained", "RoBERTa." ]

[ 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "This", "model", "assumes", "that", "there", "are", "three", "factors", "in", "emotion", "prediction:", "the", "speaker,", "the", "context", "from", "the", "preceding", "utterances", "and", "the", "emotion", "of", "the", "preceding", "utterances." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "DialogueRNN", "(Majumder", "et", "al.,", "2019)", "uses", "a", "GRU", "network", "to", "keep", "track", "of", "the", "individual", "party", "states", "in", "the", "conversation", "to", "predict", "emotions." ]

[ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "KET", "(Zhong", "et", "al.,", "2019)", "is", "a", "Knowledge", "Enriched", "Transformer", "that", "reflects", "contextual", "utterances", "with", "a", "hierarchical", "self-attention", "and", "leverages", "external", "commonsense", "knowledge", "by", "using", "a", "context-aware", "affective", "graph", "attention", "mechanism." ]

[ 1, 0, 0, 0, 0, 0, 0, 1, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "show", "that", "the", "proposed", "approach", "is", "effective", "by", "comparing", "it", "with", "various", "baselines", "and", "the", "stateof-the-art", "methods." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "4.3", "Previous", "Method" ]

[ 0, 0, 0 ]

[ "All", "experiments", "are", "conducted", "on", "one", "V100", "GPU", "with", "32GB", "memory." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "select", "the", "model", "with", "the", "best", "performance", "on", "the", "validation", "set." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "learning", "rate", "scheduler", "used", "for", "training", "is", "get_linear_schedule_with_warmup,", "and", "the", "maximum", "value", "of", "10", "is", "used", "for", "the", "gradient", "clipping." ]

[ 0, 3, 4, 4, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 3, 4 ]

[ "The", "optimizer", "is", "AdamW", "and", "the", "learning", "rate", "is", "1e-5", "as", "an", "initial", "value." ]

[ 0, 0, 0, 0, 0, 0, 3, 4, 0, 5, 0, 0, 0, 0 ]

[ "We", "use", "the", "pre-trained", "model", "from", "the", "huggingface", "library", "2." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "4.2", "Training", "Setup" ]

[ 0, 0, 0 ]

[ "Sentiment", "labels", "are", "not", "provided,", "but", "sentiment", "classes", "can", "be", "grouped", "as", "follows:", "positive:", "{joyful,", "peaceful,", "powerful},", "negative:", "{scared,", "mad,", "sad},", "neutral:", "{neutral}" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "EmoryNLP,", "like", "MELD,", "is", "also", "a", "dataset", "based", "on", "Friends", "TV", "show,", "but", "the", "emotion-inventory", "is", "given", "as", "\"joyful,", "peaceful,", "powerful,", "scared,", "mad,", "sad", "and", "neutral\"." ]

[ 13, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "MELD’s", "emotion-inventory", "is", "given", "as", "\"anger,", "disgust,", "sadness,", "joy,", "surprise,", "fear", "and", "neutrality\"", "following", "Ekman", "(Ekman,", "1992)", "and", "sentiment-inventory", "is", "given", "as", "\"positive,", "negative", "and", "neutral\"." ]

[ 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "MELD", "is", "a", "dataset", "based", "on", "Friends", "TV", "show", "and", "provides", "two", "taxonomy:", "emotion", "and", "sentiment." ]

[ 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Since", "more", "than", "82%", "of", "the", "data", "are", "tagged", "as", "neutral,", "neutral", "emotions", "are", "excluded", "when", "evaluating", "systems", "with", "Micro-F1", "as", "did", "in", "the", "previous", "studies." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0 ]

[ "DailyDialog", "is", "a", "dataset", "of", "daily", "conversations", "between", "two", "speakers", "and", "the", "emotion-inventory", "is", "given", "as", "\"anger,", "disgust,", "fear,", "joy,", "surprise,", "sadness", "and", "neutral\"." ]

[ 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "test", "dataset", "is", "a", "conversation", "involving", "two", "later", "speakers." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "train", "and", "development", "dataset", "is", "a", "conversation", "involving", "the", "previous", "eight", "speakers,", "and", "the", "train", "and", "development", "are", "divided", "into", "random", "splits", "at", "a", "ratio", "of", "9:1." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 5 ]

[ "IEMOCAP", "is", "a", "dataset", "involving", "10", "speakers,", "and", "each", "conversation", "involves", "2", "speakers", "and", "the", "emotion-inventory", "is", "given", "as", "\"happy,", "sad,", "angry,", "excited,", "frustrated", "and", "neutral\"." ]

[ 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "statistics", "of", "the", "dataset", "are", "shown", "in", "Table", "1." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "MELD", "(Poria", "et", "al.,", "2019)", "and", "EmoryNLP", "(Zahiri", "and", "Choi,", "2018)", "are", "multi-party", "datasets,", "while", "IEMOCAP", "(Busso", "et", "al.,", "2008)", "and", "DailyDialog", "(Li", "et", "al.,", "2017)", "are", "dyadic-party", "datasets." ]

[ 13, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "experiment", "on", "four", "benchmark", "datasets." ]

[ 0, 0, 0, 0, 0, 0 ]

[ "4.1", "Dataset" ]

[ 0, 0 ]

[ "4", "Experiments" ]

[ 0, 0 ]

[ "MELD", "and", "EmoryNLP", "are", "used", "to", "measure", "weighted", "avg", "F1", "for", "both", "emotion", "(7)", "and", "sentiment", "(3)", "classes." ]

[ 13, 0, 13, 0, 0, 0, 0, 7, 8, 8, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "DailyDialog", "uses", "7", "classes", "for", "training,", "but", "we", "measure", "Macro-F1", "for", "only", "6", "classes", "excluding", "neutral." ]

[ 13, 0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0 ]

[ "Dataset" ]

[ 0 ]

[ "The", "objective", "is", "to", "minimize", "the", "cross", "entropy", "loss", "so", "that", "et", "is", "the", "same", "as", "the", "ground", "truth", "emotional", "label." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "3.5.2", "Emotion", "Prediction" ]

[ 0, 0, 0 ]

[ "where,", "Wp", "is", "a", "matrix", "that", "projects", "the", "pretrained", "memory", "to", "the", "dimension", "of", "the", "context", "output,", "and", "is", "used", "only", "when", "PM", "and", "CoM", "are", "different", "pre-trained", "language", "models." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0 ]

[ "ot", "=", "ct", "+", "Wp(ktt)" ]

[ 0, 0, 0, 0, 0 ]

[ "Finally,", "the", "output", "vector", "ot", "is", "obtained", "by", "adding", "ktt", "and", "ct", "in", "Equation", "4." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "GRU", "is", "composed", "of", "2-layers,", "the", "dimension", "of", "the", "output", "vector", "is", "hc,", "and", "the", "dropout", "is", "set", "to", "0.3", "during", "training." ]

[ 0, 0, 0, 0, 0, 0, 3, 4, 4, 4, 4, 0, 3, 0, 0, 3, 0, 0, 0, 5, 0, 0 ]

[ "R1" ]

[ 0 ]

[ "We", "track", "and", "capture", "the", "sequential", "position", "information", "of", "ki", "using", "a", "unidirectional", "GRU:" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "other", "words,", "since", "the", "flow", "of", "conversation", "changes", "as", "it", "progresses,", "the", "effect", "on", "emotion", "may", "differ", "depending", "on", "the", "distance", "from", "the", "current", "utterance." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "tracking", "method", "assumes", "that", "the", "importance", "of", "all", "previous", "speaker", "utterances", "to", "the", "current", "emotion", "is", "not", "equal", "and", "varies", "with", "the", "distance", "of", "the", "current", "utterance." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "3.5.1", "Tracking", "Method", "We", "use", "ki", "tracking", "method", "using", "GRU." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Therefore,", "we", "assume", "that", "utterances", "close", "to", "the", "current", "utterance", "will", "be", "important", "in", "emotional", "recognition." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11, 12 ]

[ "In", "many", "dialogue", "systems", "(Zhang", "et", "al.,", "2018b;", "Ma", "et", "al.,", "2019),", "it", "is", "known", "that", "utterances", "close", "to", "the", "current", "turn", "are", "important", "for", "response." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "combine", "CoM", "and", "PM", "to", "predict", "the", "speaker’s", "emotion." ]

[ 0, 0, 1, 0, 1, 0, 0, 0, 0, 0 ]

[ "3.5", "CoMPM:", "Combination", "of", "CoM", "and", "PM" ]

[ 0, 1, 0, 0, 1, 0, 1 ]

[ "Wo", "and", "Wp", "are", "linear", "matrices." ]

[ 0, 0, 0, 0, 0, 0 ]

[ "A,", "B,", "and", "C", "refer", "to", "the", "participants", "in", "the", "conversation,", "where", "sA", "=", "pu1", "=", "pu3", "=", "pu6,", "sB", "=", "pu2", "=", "pu5,", "sC", "=", "pu3", "." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Since", "<cls>", "is", "mostly", "used", "for", "the", "task", "of", "classifying", "sentences,", "we", "use", "the", "embedding", "output", "of", "the", "<cls>", "token", "as", "a", "vector", "representing", "the", "utterance", "as", "follows:" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "If", "the", "speaker", "has", "never", "appeared", "before", "the", "current", "turn,", "the", "result", "of", "the", "pre-trained", "memory", "is", "considered", "a", "zero", "vector." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Inspired", "by", "previous", "studies", "that", "the", "speaker’s", "knowledge", "helps", "to", "judge", "emotions,", "we", "extract", "and", "track", "pre-trained", "memory", "from", "the", "speaker’s", "previous", "utterances", "to", "utilize", "the", "emotions", "of", "the", "current", "utterance", "ut." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Pre-trained", "language", "models", "can", "be", "trained", "on", "numerous", "corpora", "and", "be", "used", "as", "an", "external", "knowledge", "base." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "External", "knowledge", "is", "known", "to", "play", "an", "important", "role", "in", "understanding", "conversation." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "<cls>", "token", "is", "concatenated", "at", "the", "beginning", "of", "the", "input", "and", "the", "output", "of", "the", "context", "model", "is", "as", "follows:" ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "use", "the", "embedding", "of", "the", "special", "token", "<cls>", "to", "predict", "emotion." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "RoBERTa", "is", "an", "unsupervised", "pre-trained", "model", "with", "largescale", "open-domain", "corpora", "of", "unlabeled", "text." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "many", "natural", "language", "processing", "tasks,", "the", "effectiveness", "of", "the", "pre-trained", "language", "model", "has", "been", "proven,", "and", "we", "also", "set", "the", "initial", "state", "of", "the", "model", "to", "RoBERTa", "(Liu", "et", "al.,", "2019)." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "use", "an", "Transformer", "encoder", "as", "a", "context", "model." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "In", "other", "words,", "the", "same", "special", "token", "appears", "before", "the", "utterances", "of", "the", "same", "speaker." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "A", "special", "token", "<sP>", "is", "introduced", "to", "distinguish", "participants", "in", "the", "conversation", "and", "to", "handle", "the", "speaker’s", "dependency", "where", "P", "is", "the", "set", "of", "participants." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "We", "consider", "multi-party", "conversations", "where", "2", "or", "more", "speakers", "are", "involved." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "u1,", "{" ]

[ 0, 0 ]

[ "The", "previous", "utterances", "are", "and", "e6", "is", "predicted", "while", "considh6", "=", "ering", "the", "relationship", "between", "u6", "and", "h6." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "example", "in", "Figure", "2", "shows", "how", "the", "model", "predicts", "the", "emotion", "of", "u6", "uttered", "by", "sA,", "given", "a", "conversation", "of", "three", "participants", "(sA,", "sB,", "sC)." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "3.3", "CoM:", "Context", "Embedding", "Module", "The", "context", "embedding", "module", "predicts", "et", "by", "considering", "all", "of", "the", "utterances", "before", "the", "t-th", "turn", "as", "the", "dialogue", "context." ]

[ 0, 1, 1, 2, 2, 0, 1, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "combination", "of", "CoM", "and", "PM", "is", "described", "in", "Section", "4.5." ]

[ 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0 ]

[ "If", "CoM", "and", "PM", "are", "based", "on", "different", "architectures,", "CoMPM", "is", "trained", "to", "understand", "each", "other’s", "representations" ]

[ 0, 1, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0 ]

[ "Therefore,", "we", "design", "the", "PM", "to", "follow", "CoM", "so", "that", "the", "output", "representations", "of", "CoM", "and", "PM", "can", "mutually", "understand", "each", "other." ]

[ 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0 ]

[ "If", "the", "CoM", "and", "PM", "are", "based", "on", "different", "backbones,", "we", "consider", "them", "to", "be", "unaligned", "with", "respect", "to", "each", "other’s", "output", "representations." ]

[ 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "second", "one", "is", "PM", "that", "leverages", "only", "the", "speaker’s", "previous", "utterances,", "through", "which", "we", "want", "to", "reflect", "the", "speaker’s", "knowledge." ]

[ 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Therefore,", "we", "propose", "a", "context", "model", "to", "handle", "the", "relationship", "between", "the", "current", "and", "the", "previous", "utterances." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "The", "first", "is", "CoM", "which", "catches", "the", "underlying", "effect", "of", "all", "previous", "utterances", "on", "the", "current", "speaker’s", "emotions." ]

[ 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Our", "ERC", "neural", "network", "model", "is", "composed", "of", "two", "modules." ]

[ 0, 11, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "Figure", "2", "shows", "an", "overview", "of", "our", "model." ]

[ 0, 0, 0, 0, 0, 0, 0, 0 ]