Datasets:

havens2

/

naacl2022

like 0

[ "We", "cast", "the", "word-based", "text", "generation", "task", "as", "a", "Markov", "Decision", "Process", "to", "apply", "RL", "methods", "(Sutton", "et", "al.,", "1998)." ]

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

[ "Unlike", "alternative", "RL", "without", "pre-training", "approaches,", "TrufLL", "manages", "to", "ask", "meaningful", "and", "valid", "questions", "on", "large", "vocabularies,", "exhibiting", "success", "rate", "and", "language", "metrics", "close", "to", "pretrain", "models", "with", "labeled", "data,", "while", "producing", "more", "original", "language." ]

[ 0, 0, 0, 0, 0, 0, 11, 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", "here", "evaluate", "it", "on", "two", "Visual", "Question", "Generation", "(VQG)", "tasks,", "the", "synthetic", "CLEVR", "dataset", "(Johnson", "et", "al.,", "2017),", "and", "the", "natural", "language", "VQAv2", "dataset", "(Goyal", "et", "al.,", "2017)." ]

[ 0, 0, 0, 0, 0, 0, 11, 12, 12, 11, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0 ]

[ "Such", "an", "approach", "injects", "a", "generic", "prior", "linguistic", "knowledge", "into", "the", "RL", "algorithm,", "is", "usable", "on", "tasks", "lacking", "in-domain", "labeled", "data,", "and", "can", "be", "easily", "transferred", "to", "new", "RL-based", "text", "generation", "tasks." ]

[ 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, 11, 12, 0 ]

[ "At", "each", "time", "step", "of", "the", "text", "generation", "process,", "TrufLL", "truncates", "its", "effective", "action", "space", "to", "a", "small", "subset", "of", "words", "provided", "by", "a", "pretrained", "task-agnostic", "language", "model." ]

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

[ "TrufLL", "leverages", "such", "structure", "to", "drive", "the", "exploration", "of", "the", "RL-based", "language", "agent", "during", "training." ]

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

[ "Yet,", "while", "large", "and", "discrete,", "a", "language", "action", "space", "contains", "a", "specific", "structure,", "made", "of", "all", "the", "syntactical", "and", "semantics", "rules", "of", "a", "given", "language." ]

[ 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 ]

[ "While", "appealing,", "such", "an", "approach", "requires", "overcoming", "the", "hurdle", "of", "the", "combinatorial", "language", "action", "space,", "a", "vocabulary", "usually", "containing", "more", "than", "10,000", "words." ]

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

[ "In", "this", "paper,", "we", "aim", "at", "learning", "a", "conditional", "language", "model", "using", "RL", "without", "a", "pre-training", "phase,", "so", "that", "(i)", "we", "get", "free", "from", "datasets", "with", "human", "annotations,", "and", "(ii)", "we", "avoid", "the", "text", "generation", "flaws", "induced", "by", "the", "common", "methods." ]

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

[ "Besides,", "combining", "pre-training", "and", "fine-tuning", "phases", "either", "barely", "change", "the", "policy", "distribution,", "or", "induces", "language", "drift", "(Lazaridou", "et", "al.,", "2020;", "Lu", "et", "al.,", "2020b),", "i.e", "the", "generated", "language", "drifts", "semantically", "or", "syntactically", "from", "natural", "language." ]

[ 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, 0, 0, 0 ]

[ "So", "far,", "RL-based", "text-generation", "tasks", "have", "relied", "on", "a", "pre-training", "phase", "to", "ease", "learning:", "the", "policy", "language", "model", "is", "trained", "with", "SL", "on", "the", "task", "dataset,", "before", "being", "fine-tuned", "with", "policy", "gradient", "methods", "(Sutton", "et", "al.,", "1999)", "on", "the", "task", "at", "hand." ]

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

[ "On", "the", "other", "hand,", "text", "generation", "can", "be", "naturally", "framed", "as", "a", "sequential", "decision", "making", "problem,", "with", "the", "sequence", "of", "words", "seen", "as", "successive", "actions", "over", "a", "vocabulary." ]

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

[ "If", "both", "answers", "match,", "the", "agent", "is", "rewarded." ]

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

[ "In", "a", "VQG", "training", "loop,", "the", "agent", "generates", "a", "question", "given", "an", "image-answer", "pair,", "which", "is", "then", "fed", "to", "a", "VQA", "model", "predicting", "an", "expected", "answer." ]

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

[ "In", "a", "conditional", "language", "generation", "task", "as", "VQG,", "TrufLL", "truncates", "the", "vocabulary", "space", "by", "using", "a", "language", "model." ]

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

[ "1Code" ]

[ 0 ]

[ "The", "pretrained", "Language", "Model", "(LM)", "is", "fine-tuned", "on", "the", "downstream", "task", "using", "a", "standard", "supervised", "learning", "(SL)" ]

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

[ "Brown", "et", "al.,", "2020),", "state-of-the", "art", "language", "processing", "systems", "rely", "on", "sequential", "transfer", "learning", "(Ruder,", "2019)." ]

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

[ "Since", "the", "development", "of", "generic", "language", "models", "trained", "on", "massive", "unlabelled", "text", "corpora", "(Radford", "et", "al.,", "2019;" ]

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

[ "1", "Introduction" ]

[ 0, 0 ]

[ "1" ]

[ 0 ]

[ "To", "our", "knowledge,", "it", "is", "the", "first", "approach", "that", "successfully", "learns", "a", "language", "generation", "policy", "without", "pre-training,", "using", "only", "reinforcement", "learning." ]

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

[ "We", "evaluate", "TrufLL", "on", "two", "visual", "question", "generation", "tasks,", "for", "which", "we", "report", "positive", "results", "over", "performance", "and", "language", "metrics,", "which", "we", "then", "corroborate", "with", "a", "human", "evaluation." ]

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

[ "Interestingly,", "this", "approach", "avoids", "the", "dependency", "to", "labelled", "datasets", "and", "inherently", "reduces", "pretrained", "policy", "flaws", "such", "as", "language", "or", "exposure", "biases." ]

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

[ "TrufLL", "thus", "enables", "to", "train", "a", "language", "agent", "by", "solely", "interacting", "with", "its", "environment", "without", "any", "task-specific", "prior", "knowledge;", "it", "is", "only", "guided", "with", "a", "task-agnostic", "language", "model." ]

[ 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 ]

[ "As", "RL", "methods", "unsuccessfully", "scale", "to", "large", "action", "spaces,", "we", "dynamically", "truncate", "the", "vocabulary", "space", "using", "a", "generic", "language", "model." ]

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

[ "This", "paper", "introduces", "TRUncated", "ReinForcement", "Learning", "for", "Language", "(TrufLL),", "an", "original", "approach", "to", "train", "conditional", "language", "models", "without", "a", "supervised", "learning", "phase,", "by", "only", "using", "reinforcement", "learning", "(RL)." ]

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

[ "Abstract" ]

[ 0 ]

[ "Florian", "Strub", "Deep", "Mind" ]

[ 0, 0, 0, 0 ]

[ "Sylvain", "Le", "Corff", "Samovar,", "Télécom", "SudParis,", "Institut", "Polytechnique", "de", "Paris,", "Palaiseau" ]

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

[ "Guillaume", "Quispe", "and", "Charles", "Ollion", "CMAP,", "École", "Polytechnique,", "Institut", "Polytechnique", "de", "Paris,", "Palaiseau" ]

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

[ "Alice", "Martin", "Samovar,", "Télécom", "SudParis,", "Institut", "Polytechnique", "de", "Paris,", "Palaiseau", "CMAP,", "École", "Polytechnique,", "Institut", "Polytechnique", "de", "Paris,", "Palaiseau" ]

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

[ "Learning", "Natural", "Language", "Generation", "with", "Truncated", "Reinforcement", "Learning" ]

[ 0, 11, 12, 12, 0, 1, 2, 2 ]

[ "12", "Proceedings", "of", "the", "2022", "Conference", "of", "the", "North", "American", "Chapter", "of", "the", "Association", "for", "Computational", "Linguistics:", "Human", "Language", "Technologies,", "pages", "12", "-", "37", "July", "10-15,", "2022", "©2022", "Association", "for", "Computational", "Linguistics" ]

[ 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 ]

[ "References" ]

[ 0 ]

[ "This", "work", "was", "supported", "in", "part", "by", "the", "Shandong", "Provincial", "Natural", "Science", "Foundation", "(No.ZR2020MF149),", "in", "part", "by", "the", "Science", "and", "Technology", "Commission", "of", "Shanghai", "Municipality", "(21511100302)." ]

[ 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 ]

[ "Acknowledgements" ]

[ 0 ]

[ "Experimental", "results", "demonstrate", "the", "effectiveness", "of", "our", "approach", "on", "three", "public", "datasets." ]

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

[ "Consequently,", "we", "combine", "attention", "score", "matrices", "with", "syntactic", "mask", "matrices", "to", "fuse", "the", "semantic", "and", "syntactic", "information." ]

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

[ "Furthermore,", "we", "construct", "syntactic", "mask", "matrices", "of", "a", "sentence", "according", "to", "the", "different", "syntactic", "distances", "that", "learn", "local", "to", "global", "structure", "information." ]

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

[ "Then,", "we", "combine", "it", "with", "the", "self-attention", "to", "compose", "the", "attention", "layer." ]

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

[ "Specifically,", "we", "first", "design", "an", "aspect-aware", "attention", "mechanism,", "which", "is", "responsible", "for", "learning", "aspectrelated", "semantic", "information." ]

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

[ "In", "this", "paper,", "we", "propose", "a", "SSEGCN", "architecture", "which", "integrates", "semantic", "information", "along", "with", "the", "syntactic", "structure", "for", "ABSA", "task." ]

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

[ "5", "Conclusion" ]

[ 0, 0 ]

[ "the", "introduction", "of", "noise", "and", "decline", "of", "the", "model", "performance." ]

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

[ "When", "the", "syntactic", "distance", "is", "greater", "than", "five,", "multiple", "syntactic", "mask", "matrices", "are", "fully", "connected", "matrices,", "and", "leads", "to" ]

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

[ "mask", "matrices", "from", "5", "to", "7", "leads", "to", "the", "performance", "degradation", "of", "SSEGCN." ]

[ 0, 0, 0, 5, 0, 5, 0, 0, 0, 0, 0, 0, 1 ]

[ "If", "you", "are", "a", "Tequila", "fan", "you", "will", "not", "be", "disappointed." ]

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

[ "Not", "as", "fact", "as", "I", "would", "have", "expect", "for", "an", "i5" ]

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

[ "The", "settings", "are", "not", "user-friendly", "either." ]

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

[ "Biggest", "complaint", "is", "Windows", "8." ]

[ 0, 0, 0, 0, 0 ]

[ "However,", "increasing", "the", "number", "of", "syntactic" ]

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

[ "One", "main", "reason", "may", "be", "that", "SSEGCN", "can", "learn", "structure", "information", "from", "local", "to", "global", "when", "the", "syntactic", "distance", "becomes", "larger", "and", "SSEGCN", "achieves", "remarkable", "results", "on", "five", "syntactic", "mask", "matrices." ]

[ 0, 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, 0, 0, 0 ]

[ "We", "observe", "that", "the", "proposed", "SSEGCN", "shows", "an", "upward", "trend", "with", "the", "increase", "of", "the", "number", "when", "the", "number", "of", "syntactic", "mask", "matrices", "is", "less", "than", "5." ]

[ 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 4, 4, 4, 4, 0, 0, 5, 6 ]

[ "We", "conduct", "different", "numbers", "of", "the", "syntactic", "mask", "matrices", "from", "1", "to", "7", "and", "the", "results", "are", "demonstrated", "in", "Figure", "5." ]

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

[ "In", "this", "section,", "we", "further", "analyze", "the", "effect", "of", "the", "multiple", "different", "syntactic", "mask", "matrices", "on", "the", "performance", "of", "SSEGCN", "in", "Restaurant", "and", "Laptop", "datasets." ]

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

[ "4.9", "Effect", "of", "Syntax-Mask" ]

[ 0, 0, 0, 0 ]

[ "Second,", "node", "representation", "will", "be", "over-smooth", "and", "obtain", "more", "redundancy", "information", "when", "the", "number", "of", "GCN", "layers", "is", "large,", "thus", "making", "model", "in", "poor", "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 ]

[ "of", "GCN", "layer", "is", "set", "to", "1,", "SSEGCN", "can", "only", "learn", "local", "node", "information", "with", "syntactic", "distance", "of", "1." ]

[ 0, 3, 4, 0, 0, 0, 5, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "First,", "if", "the", "number" ]

[ 0, 0, 0, 0 ]

[ "As", "shown", "in", "Figure", "4,", "experimental", "results", "show", "that", "our", "model", "achieves", "the", "best", "performance", "with", "2", "layers." ]

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

[ "In", "this", "section,", "we", "investigate", "the", "effect", "of", "the", "layer", "number", "ranging", "from", "1", "to", "5", "on", "the", "Restaurant", "and", "Laptop", "datasets." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 4, 0, 0, 5, 0, 5, 0, 0, 13, 0, 13, 0 ]

[ "4.8", "Effect", "of", "SSEGCN", "Layers" ]

[ 0, 0, 0, 1, 0 ]

[ "Thus,", "SSEGCN", "can", "accurately", "find", "the", "opinion", "words", "corresponding", "to", "each", "aspect", "term." ]

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

[ "For", "sentence", "“great", "food", "but", "the", "service", "was", "dreadful!”,", "our", "model", "considers", "the", "aspect-related", "semantic", "correlations", "by", "introducing", "aspect-aware", "attention", "and", "combining", "syntactic", "mask", "matrices." ]

[ 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 ]

[ "For", "the", "harder", "example", "of", "multiple", "aspects", "with", "different", "sentiment", "polarities,", "our", "model", "also", "performs", "well." ]

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

[ "Our", "SSEGCN", "model", "considers", "the", "semantics", "of", "the", "sentence", "and", "reduce", "the", "attention", "weight", "on", "the", "word", "“friendly”", "through", "syntactic", "distance", "mask", "and", "aspect-aware", "attention." ]

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

[ "“The", "staff", "should", "be", "a", "bit", "more", "friendly.”,", "our", "model", "correctly", "identifies", "the", "sentiment", "of", "aspect", "term", "“staff”", "as", "negative." ]

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

[ "For", "the", "sentence" ]

[ 0, 0, 0 ]

[ "To", "further", "demonstrate", "how", "our", "SSEGCN", "improves", "ABSA", "task,", "we", "use", "the", "two", "test", "examples", "in", "Figure", "3", "to", "visualize", "attention", "scores." ]

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

[ "4.7", "Visualization" ]

[ 0, 0 ]

[ "Additionally,", "when", "dealing", "with", "complex", "sentences", "with", "implicit", "sentiment", "expressions,", "our", "SSEGCN", "can", "achieve", "better", "performance." ]

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

[ "Our", "SSEGCN", "correctly", "predicts", "all", "the", "samples,", "and", "the", "results", "suggest", "that", "SSEGCN", "effectively", "combines", "syntactic", "structure", "and", "semantic", "information." ]

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

[ "Thus,", "the", "ability", "to", "learn", "integral", "semantics", "of", "a", "sentence", "is", "a", "significant", "factor", "for", "ABSA", "task." ]

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

[ "For", "the", "sentence", "“Not", "as", "fact", "as", "I", "would", "have", "expect", "for", "an", "i5”,", "CDT", "does", "not", "obtain", "the", "representation", "of", "the", "keyword", "“not”", "from", "a", "syntactic", "point", "of", "view", "and", "thus", "produces", "wrong", "prediction." ]

[ 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, 0, 0, 0 ]

[ "The", "last", "two", "examples", "have", "no", "explicit", "sentiment", "expression." ]

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

[ "In", "the", "third", "sample,", "the", "key", "point", "is", "capturing", "the", "negated", "semantics", "which", "is", "most", "methods", "tend", "to", "ignore", "and", "easily", "make", "wrong", "predictions." ]

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

[ "has", "the", "interfering", "word", "“Biggest”,", "which", "may", "neutralize", "the", "polarity", "of", "the", "word", "“complaint”." ]

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

[ "The", "second", "sample", "“Biggest", "complaint", "is", "Windows", "8.”" ]

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

[ "has", "two", "aspects", "(“food”", "and", "“service”)", "with", "contrast", "sentiment", "polarities,", "which", "may", "interfere", "with", "the", "prediction", "of", "the", "attention", "models." ]

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

[ "The", "first", "sample", "“great", "food", "but", "the", "service", "was", "dreadful!”" ]

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

[ "The", "notations", "P,", "N", "and", "O", "in", "the", "table", "represent", "positive,", "negative", "and", "neutral", "sentiment,", "respectively." ]

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

[ "Particularly,", "we", "compare", "SSEGCN", "with", "ATAE-LSTM,", "IAN", "and", "CDT", "in", "Table", "4", "which", "contain", "their", "predictions", "and", "the", "corresponding", "truth", "labels", "on", "these", "sentences." ]

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

[ "To", "examine", "whether", "SSEGCN", "is", "able", "to", "capture", "syntax", "and", "semantic", "information", "for", "improving", "ABSA,", "we", "conduct", "case", "study", "with", "a", "few", "of", "sample", "sentences." ]

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

[ "4.6", "Case", "Study" ]

[ 0, 0, 0 ]

[ "In", "summary,", "the", "ablation", "experimental", "results", "show", "that", "each", "component", "contributes", "to", "our", "entire", "model." ]

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

[ "In", "addition,", "the", "removal", "of", "syntactic", "mask", "matrix", "and", "aspect-aware", "attention", "leads", "to", "a", "significant", "performance", "drop,", "which", "further", "indicates", "that", "these", "two", "components", "play", "crucial", "roles", "in", "SSEGCN", "for", "ABSA", "task." ]

[ 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, 1, 0, 11, 0 ]

[ "Second,", "removing", "syntactic", "mask", "matrix", "leads", "to", "dropping", "0.90%,", "1.42%", "and", "0.88%", "in", "accuracy", "on", "Restaurant,", "Laptop", "and", "Twitter", "respectively,", "which", "indicates", "that", "syntactic", "mask", "matrix", "can", "assist", "GCNs", "to", "learn", "better", "syntactic", "structure", "information", "in", "original", "dependency", "trees." ]

[ 0, 0, 0, 0, 0, 0, 0, 0, 9, 9, 0, 9, 0, 7, 0, 13, 13, 0, 13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]

[ "tion", "between", "aspect", "and", "contextual", "words." ]

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

[ "It", "indicates", "that", "aspect-aware", "attention", "is", "essential", "to", "capture", "the", "correlated", "semantic", "informa" ]

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

[ "We", "can", "also", "notice", "that", "model", "without", "aspect-aware", "attention", "performs", "unsatisfactory,", "which", "indicates", "that", "the", "model", "lacks", "of", "the", "ability", "to", "capture", "aspect-related", "semantics,", "resulting", "in", "1.70%,", "1.58%", "and", "1.47%", "reductions", "in", "accuracy", "on", "Restaurant,", "Laptop", "and", "Twitter,", "respectively." ]

[ 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, 9, 9, 0, 9, 0, 0, 7, 0, 13, 13, 0, 13, 0 ]

[ "First,", "we", "observe", "that", "removal", "of", "self-attention", "degrades", "the", "performance,", "verifying", "that", "global", "semantics", "of", "the", "sentence", "is", "necessary", "for", "ABSA." ]

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

[ "The", "basic", "SSEGCN", "is", "regarded", "as", "a", "baseline", "model." ]

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

[ "As", "illustrated", "in", "Table", "3", ",", "we", "further", "conduct", "an", "ablation", "study", "to", "examine", "the", "effectiveness", "of", "different", "modules", "in", "SSEGCN." ]

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

[ "4.5", "Ablation", "Study" ]

[ 0, 0, 0 ]

[ "Combining", "our", "SSEGCN", "with", "BERT,", "the", "results", "show", "that", "the", "effectiveness", "of", "this", "powerful", "model", "is", "further", "improved,", "justifying", "that", "SSEGCN", "learns", "more", "syntactic", "and", "semantic", "knowledge", "can", "empower", "ABSA." ]

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

[ "On", "the", "other", "hand,", "we", "can", "observe", "that", "the", "basic", "BERT", "has", "been", "significantly", "better", "than", "most", "ABSA", "models." ]

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

[ "In", "GCN-based", "models,", "our", "SSEGCN", "learns", "structure", "information", "from", "local", "to", "global", "and", "considers", "aspect-related", "semantic", "information,", "and", "performs", "significantly", "better", "than", "the", "previous", "GCN-based", "models", "(i.e.,", "CDT,", "TD-GAT,", "BiGCN,", "KuamGCN,", "R-GAT,", "DGEDT", "and", "DualGCN)", "that", "verifies", "the", "effectiveness", "of", "fusing", "syntactic", "and", "semantic", "information." ]

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

[ "In", "particular,", "GCN-based", "models", "take", "into", "account", "the", "syntactic", "structure", "of", "a", "sentence", "and", "capture", "longterm", "syntactic", "dependencies", "between", "the", "aspect", "word", "and", "the", "opinion", "word,", "hence", "outperform", "all", "attention-based", "methods." ]

[ 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 ]

[ "Experimental", "results", "show", "that", "our", "SSEGCN", "model", "achieves", "best", "performance", "on", "the", "three", "datasets." ]

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