| """Evidence-based, multi-label research topics, subtopics and independent methods. |
| |
| This module only classifies already admitted finance papers. It never decides |
| whether a paper belongs in the collection. Rules and manual overrides are saved |
| with matching source passages so classifications can be reviewed and revised. |
| """ |
| import re |
| import json |
|
|
| VERSION = '2.0' |
|
|
| |
| TOPICS = [ |
| ('alpha_mining','Alpha Mining / 量化因子挖掘','Alpha Mining','因子表达式搜索、自动化 alpha 发现与因子评价；不把一般 alpha 超参数或资产定价混入。'), |
| ('time_series','Time Series / 时间序列','Time Series','有金融应用证据的序列预测、生成、表征、异常、缺失及动态建模。'), |
| ('asset_pricing','Asset Pricing / 资产定价','Asset Pricing','收益横截面、风险溢价、随机折现因子与资产定价模型。'), |
| ('derivatives','Derivatives & Hedging / 衍生品定价与对冲','Derivatives','期权、隐含波动率曲面、衍生品定价与对冲。'), |
| ('portfolio','Portfolio & Investment / 投资组合与投资管理','Portfolio','资产配置、组合选择、投资评价与有效前沿；通用模型组合不据此归类。'), |
| ('trading','Trading & Microstructure / 交易与市场微观结构','Trading','交易策略、执行、做市、订单簿、市场仿真与预测市场。'), |
| ('market_prediction','Market Prediction / 股价、收益与波动率预测','Market Prediction','股票方向与收益、波动率、汇率或加密资产市场预测。'), |
| ('credit_risk','Credit & Financial Risk / 信贷与金融风险','Credit and Risk','授信、违约、信用风险、运营风险和尾部风险。'), |
| ('insurance','Insurance & Actuarial / 保险与精算','Insurance','承保、保费、理赔、保险业务数据与精算应用。'), |
| ('fraud_aml','Fraud & AML / 欺诈检测与反洗钱','Fraud and AML','支付和金融交易欺诈、反洗钱、金融诈骗。'), |
| ('financial_nlp','Financial NLP / 金融文本、情感与事件','Financial NLP','金融文本情感、立场、信息抽取、生成和摘要。'), |
| ('financial_qa','Financial QA & Reasoning / 金融问答与数值推理','Financial QA','金融知识问答、财报表格计算、数学和多跳推理。'), |
| ('financial_ir','Financial IR & RAG / 金融检索与RAG','Financial IR and RAG','金融信息检索、FiQA 检索评测、嵌入、重排与金融 RAG。'), |
| ('document_ai','Document AI / 票据、财报与图表理解','Document AI','票据发票、OCR、文档版面、财务表格与金融图表；普通表格分类另标技术方法。'), |
| ('financial_agents','Financial Agents / 金融Agent与决策','Financial Agents','金融研究、投资分析、金融工具使用和金融决策 Agent。'), |
| ('macro_economics','Macro & Economic Forecasting / 宏观经济与预测','Macro and Economics','宏观指标、货币政策、经济预测、经济活动仿真。'), |
| ('corporate_finance','Corporate Finance & Accounting / 公司金融与会计','Corporate and Accounting','盈利、披露、并购、财务报告、会计审计与税务。'), |
| ('esg','ESG & Sustainable Finance / ESG与可持续金融','ESG','企业可持续披露、绿色投资、ESG 分析和合规审计。'), |
| ('banking','Banking & Customer Service / 银行业务与客户服务','Banking','银行营销、客户意图、客服对话、债务催收及金融服务。'), |
| ('blockchain','Blockchain & Digital Assets / 区块链与数字资产','Blockchain','比特币、以太坊、链上交易、智能合约及加密资产数据。'), |
| ('financial_safety','Financial Safety & Compliance / 金融安全与合规','Safety and Compliance','金融隐私、金融 LLM 安全、投资建议风险、监管及可信评测。'), |
| ('financial_models','Financial Models & Benchmarks / 金融模型与评测资源','Models and Benchmarks','金融领域预训练、领域适配、跨任务金融评测与综合金融资源。'), |
| ('other_finance','Other Financial Applications / 其他金融应用','Other Finance','金融相关性已确认，但当前证据未支持更具体的上述任务分类。'), |
| ] |
| TOPIC_BY_ID={x[0]:dict(id=x[0],label=x[1],sheet=x[2],description=x[3]) for x in TOPICS} |
|
|
| |
| |
| RULES = [ |
| ('alpha_mining','factor_discovery','Alpha Discovery / Alpha与因子发现',r'\balpha (?:factor|mining|discovery)|formulaic alpha|alpha(?:bench|sage|probe|pareto|qcm)\b|quantitative factor (?:mining|discovery)|financial factor (?:mining|discovery)|Cognitive Alpha'), |
| ('alpha_mining','factor_evaluation','Factor Evaluation / 因子评价与回测',r'factor (?:evaluat|screen|select)|alpha (?:evaluat|benchmark)|backtesting.*alpha|alpha.*backtesting'), |
| ('time_series','ts_forecasting','Forecasting / 序列预测',r'time[- ]?series.{0,75}forecast|forecast.{0,65}time[- ]?series|\b(?:exchange[- ]?rate|fred-md|M4[- ](?:Yearly|Quarterly|Monthly|Daily|Weekly))\b|\b(?:TimeGrad|FEDformer|Autoformer|iTransformer|TimesNet|TimeMixer|N-BEATS)\b'), |
| ('time_series','ts_foundation','Representation & Foundation Models / 表征与基础模型',r'time[- ]?series.{0,65}(?:foundation|pretrain|pre-train|representation|embedding)|(?:foundation|pretrain|pre-train|representation).{0,65}time[- ]?series|\b(?:Moirai|Time-LLM|Chronos|Time-MoE)\b'), |
| ('time_series','ts_generation','Generation & Simulation / 序列生成与仿真',r'time[- ]?series.{0,65}(?:generat|synthes|simulat)|(?:generat|synthes|simulat).{0,65}time[- ]?series|\b(?:CTBench|SigDiffusions|Diffusion-TS|TimeTok)\b'), |
| ('time_series','ts_anomaly','Anomaly Detection / 序列异常检测',r'time[- ]?series.{0,65}(?:anomal|outlier)|(?:anomal|outlier).{0,65}time[- ]?series|\bNIPS-TS-CREDITCARD\b'), |
| ('time_series','ts_imputation','Imputation & Irregular Series / 缺失与不规则序列',r'time[- ]?series.{0,80}(?:imput|irregular|missing)|(?:imput|irregular|missing).{0,80}time[- ]?series'), |
| ('time_series','ts_dynamics','Causality & Dynamics / 序列因果与动力学',r'(?:causal|differential equation|dynamics|point process).{0,75}(?:time[- ]?series|stochastic process|stock|financ)|time[- ]?series.{0,75}(?:causal|dynamics)|temporal point process|stochastic differential equation|\bNeural MJD\b'), |
| ('time_series','ts_uncertainty','Probabilistic Forecasting & Calibration / 概率预测与校准',r'(?:probabilistic|conformal|uncertainty).{0,65}(?:forecast|time[- ]?series)|(?:forecast|time[- ]?series).{0,65}(?:probabilistic|conformal|uncertainty)'), |
| ('time_series','ts_general','General Time-Series Methods / 综合序列方法',r'time[- ]?series|\b(?:NASDAQ|NYSE|SP500|S&P 500)\b.{0,50}(?:dataset|forecast|stock data)'), |
| ('asset_pricing','pricing_models','Pricing Models & Risk Premia / 定价模型与风险溢价',r'asset pricing|risk premi(?:um|a)|stochastic discount factor|cross[- ]sectional.{0,25}(?:asset|return)|lookahead factor models|factor models for quantitative investing'), |
| ('derivatives','option_pricing','Option Pricing & PDEs / 期权定价与PDE',r'option pric|Black[- ]Scholes|American option|derivative pric|pricing.{0,20}derivatives'), |
| ('derivatives','hedging','Hedging / 对冲',r'\bhedging\b|hedge.{0,25}(?:option|derivative|asset|financial)|(?:option|financial).{0,25}hedge'), |
| ('derivatives','volatility_surface','Implied Volatility / 隐含波动率曲面',r'implied volatility|volatility surface'), |
| ('portfolio','allocation','Asset Allocation / 资产配置',r'asset allocation|asset allocat|Black[- ]Litterman|investment polic|sustainable investment'), |
| ('portfolio','selection','Portfolio Selection & Optimization / 组合选择与优化',r'portfolio[- ](?:optimi|alloc|select|manag|invest|return|value|benchmark)|efficient frontier|Sharpe ratio|stochastic discount factor portfolios'), |
| ('trading','trading_execution','Trading & Execution / 交易策略与执行',r'\btrading\b|trade execution|optimal execution|\bCryptoTrade\b|\bTrade in Minutes\b|arbitrage|financial backtest'), |
| ('trading','microstructure','Order Books & Market Impact / 订单簿与市场冲击',r'order book|market impact|market microstructure|market making|market maker|trading.volume|\bLOB[- ]Bench\b'), |
| ('trading','market_simulation','Market Simulation / 金融市场仿真',r'(?:financial|stock) market simulat|simulat.{0,30}(?:stock|financial) market|\b(?:MarketSim|TwinMarket|MarS:)\b|financial world model'), |
| ('trading','prediction_markets','Prediction Markets / 预测市场',r'prediction markets?'), |
| ('market_prediction','stock_prediction','Stock Movement & Returns / 股价方向与收益预测',r'(?:stock|equity|equities).{0,60}(?:predict|forecast)|(?:predict|forecast).{0,60}(?:stock|equity|equities)|financial forecast|stocknet|causalstock|market comments from stock'), |
| ('market_prediction','volatility_prediction','Volatility & Tail Forecasts / 波动率与尾部预测',r'volatility (?:predict|forecast)|(?:predict|forecast).{0,55}volatility|financial asset tail dynamics|asset.{0,30}tail dynamics|extremal dependence.{0,30}financial assets'), |
| ('market_prediction','fx_crypto','FX & Crypto Forecasts / 汇率与加密资产预测',r'exchange[- ]?rate|(?:crypto|bitcoin).{0,65}(?:forecast|predict)|(?:forecast|predict).{0,65}(?:crypto|bitcoin)'), |
| ('credit_risk','credit_default','Credit Scoring & Default / 信用评分与违约',r'credit[- _](?:risk|scor|default|approval|screen|g\b|reg\b)|German[- ]?Credit|Australian[- ]Credit|\b(?:HELOC|FICO|GiveMeSomeCredit|HomeCredit)\b|home[- ]credit|loan[- ](?:default|approv|predict|decision|data)|default.of.credit|Taiwan(?:ese)? credit|loan problem|credit dataset|credit card clients'), |
| ('credit_risk','financial_risk','Financial Risk / 金融风险管理',r'financial risk|operational risk|financial.{0,25}stress test|financial asset tail|creditworthiness|credit[- ](?:card|payment).{0,35}default'), |
| ('insurance','insurance_business','Insurance & Claims / 承保与理赔',r'insurance|actuarial|allstate|healthcare claims|health claims|inslogic|Insur(?:ance)?QA|homesite|insurance claims'), |
| ('fraud_aml','financial_fraud','Payment & Transaction Fraud / 支付与交易欺诈',r'(?:credit[- ]?card|financial|insurance|payment|transaction).{0,65}(?:fraud|anomal)|(?:fraud|anomal).{0,65}(?:credit[- ]?card|financial|insurance|payment|transaction)|\b(?:PaySim|BankSim|T-Finance)\b|innovative image fraud|identity fraud via dialogue|financial scam|procurement fraud'), |
| ('fraud_aml','aml','Anti-Money Laundering / 反洗钱',r'money[- ]launder|anti[- ]money|\b(?:AMLSim|Elliptic dataset)\b|illicit bitcoin|bitcoin.{0,55}illicit'), |
| ('fraud_aml','scam','Scams & Financial Misinformation / 诈骗与金融虚假信息',r'\bscam|fraud detection|financial misinformation|counterfactual financial misinformation|financial.{0,35}misinformation'), |
| ('financial_nlp','sentiment_stance','Sentiment & Stance / 情感与立场',r'financial[- ]?phrase[- ]?bank|financialphrasebank|financial sentiment|sentiment.{0,60}(?:financ|stock|market)|(?:financ|stock|market).{0,60}(?:sentiment|stance)|\bFinEntity\b|\bEFSA\b'), |
| ('financial_nlp','information_extraction','Entities, Events & Causality / 实体、事件与关系抽取',r'financial.{0,70}(?:event extract|entity|relation|causal)|(?:entity|event|relation).{0,45}financial|\b(?:(?-i:FiNER)|FinCausal|FinRE|ChFinAnn|DuEE-Fin|Doc2EDAG|FinKario)\b'), |
| ('financial_nlp','text_generation','Generation & Summarization / 金融文本生成与摘要',r'\bECTSum\b|earnings? call.{0,65}summari|summari.{0,65}(?:financ|earnings?|report)|financial report generation|market comments|financial text generation|accounting audit procedures generation'), |
| ('financial_qa','numerical_reasoning','Numerical & Table Reasoning / 数值与表格推理',r'\b(?:FinQA|ConvFinQA|TAT[- ]?QA|TAT[- ]?DQA|MultiHiertt|FinanceMATH|BizBench|DocFinQA|FinChain|FinanceReasoning|FinMME|CFMME)\b|financial (?:numerical |mathematical |multi-hop |table[- ]text )?reasoning|(?:numerical|math).{0,65}financ'), |
| ('financial_qa','knowledge_qa','Knowledge & Conversational QA / 知识与对话问答',r'\b(?:FinTextQA|FinanceBench|PACIFIC|SustainableQA|SAHM)\b|financial.{0,50}question answer|question answer.{0,55}financ|financial.{0,35}multi-hop'), |
| ('financial_ir','retrieval','Retrieval, Embedding & Reranking / 检索、嵌入与重排',r'\bF\s*I\s*QA\b|\bFinMTEB\b|financial.{0,65}\b(?:retriev\w*|search|text embedding|rerank\w*)|\b(?:retriev\w*|search|text embedding).{0,65}\bfinanc\w*|\bFinSearchComp\b'), |
| ('financial_ir','rag','Financial RAG & Long Documents / 金融RAG与长文档',r'\b(?:FinRAGBench|FinRAGBench-V|REAL-MM-RAG|OmniEval)\b|financial.{0,65}(?:RAG|retrieval.augmented)|(?:RAG|retrieval.augmented).{0,65}financ'), |
| ('document_ai','receipts_invoices','Receipts, Invoices & OCR / 票据发票与OCR',r'\b(?:SROIE|CORD|Kleister|Chargrid)\b|receipts?|invoices?|financial.{0,30}(?:OCR|document pars)'), |
| ('document_ai','financial_charts','Charts & Visual Understanding / 金融图表与视觉理解',r'financial.{0,55}(?:chart|visual)|(?:chart|visual).{0,55}financ|\b(?:FinChart|FinMME|CFMME|VisFinEval|FCMR|FinRAGBench-V|ChartVA)\b'), |
| ('document_ai','reports_tables','Report Layout & Structured Extraction / 财报版面与结构提取',r'\b(?:XBRL|EDINET-Bench)\b|table of contents.{0,60}(?:financial|ESG)|(?:financial|report).{0,50}(?:layout|table extraction|document understanding|parsing)|document understanding.{0,60}(?:report|financial)'), |
| ('financial_agents','agent_decisions','Investment & Financial Research Agents / 投研与金融决策Agent',r'(?:financial|invest(?:ment|or)|trading|quantitative|ESG).{0,60}\bagent|\bagent.{0,65}(?:financial|investment|trading)|\b(?:FinCon|FinSight|CryptoTrade|SocioDojo|INVESTORBENCH|FinDeepForecast|MoCA-Agent)\b'), |
| ('financial_agents','tools_research','Financial Tools & Deep Research / 金融工具与深度研究',r'financial.{0,45}(?:deep research|tool|search and reasoning)|FinSearchComp|MoCA-Agent|financial report generation'), |
| ('macro_economics','macro_forecasting','Macro Indicators & Policy / 宏观指标与政策',r'macroeconomic|\b(?:FRED[- ]MD|FOMC)\b|monetary polic|fiscal polic|economic (?:growth|indicator|forecast)|unemployment rate|Trillion Dollar Words|country.level socio.economic indicators'), |
| ('macro_economics','economic_simulation','Economic Behavior & Simulation / 经济行为与仿真',r'EconAgent|economic risk preferences|simulat.{0,40}(?:economic|economy)|economic.{0,35}(?:simulat|interactions)|economy of minds'), |
| ('corporate_finance','earnings_disclosure','Earnings & Corporate Disclosures / 盈利与公司披露',r'earnings? call|earning surprise|earnings? forecast|company.{0,20}(?:earnings|revenue)|revenue forecast|corporate.{0,30}(?:disclosure|financial)|financial disclosure|\b(?:EDGAR|EDINET|ECTSum|FinCall-Surprise|(?-i:CAME))\b|SEC filings|financial statements?|annual reports?'), |
| ('corporate_finance','ma','Mergers & Acquisitions / 并购与公司事件',r'merger|acquisition.{0,35}financial|M&A|merger.arbitrage'), |
| ('corporate_finance','accounting_tax','Accounting, Audit & Tax / 会计、审计与税务',r'\btax(?:ation|payer| returns?| practice| planning| compliance)|TaxPraBen|accounting (?:audit|report|standard|practice|procedure)|audit procedures|bookkeeping|tax and financial education'), |
| ('esg','esg_disclosure','ESG Disclosure & Greenwashing / ESG披露与漂绿',r'\bESG\b|sustainability|greenwashing|environmental claim|corporate promise|sustainableqa|ESGenius|ESGPT'), |
| ('esg','sustainable_investment','Sustainable Investment / 可持续投资',r'sustainable investment|climate investment|InvestESG|green financ|sustainable financ|environmental development projects'), |
| ('banking','customer_intents','Banking Dialogue & Intent / 银行对话与意图',r'\bBANKING77\b|banking|debt collection|bank.{0,30}(?:customer|dialogue|intent|service)|\bBQ Corpus\b|personal finance'), |
| ('banking','bank_marketing','Bank Marketing & Churn / 银行营销与客户流失',r'bank[- _]marketing|bank marketing|bank_customer_churn|bank customer churn|term[- ]deposit'), |
| ('blockchain','crypto_transactions','Crypto & Blockchain Transactions / 加密资产与链上交易',r'bitcoin|ethereum|cryptocurrenc|blockchain|\b(?:CTBench|CryptoTrade|EX-Graph|BlockScan)\b'), |
| ('blockchain','smart_contracts','Smart Contracts & DeFi / 智能合约与DeFi',r'smart contract|\b(?:EVMbench|EVM-QuestBench|DeFi)\b|sandwich equilibria|sandwich attack'), |
| ('financial_safety','privacy_security','Financial Privacy & Security / 金融隐私与信息安全',r'financial.{0,65}(?:privacy|security|red.team|harm|safety)|(?:privacy|security|red.team|harm).{0,65}financ|deanonymization.{0,40}bitcoin|credit card (?:numbers?|details|information)|bank card.{0,30}password|wrong credit card|incorrect credit card'), |
| ('financial_safety','trust_compliance','Trust, Compliance & Bias / 可信、合规与偏差审计',r'\bFinTrust\b|Shari|financial.{0,65}(?:compliance|trustworthiness|look.ahead|bias|abstention)|(?:look.ahead|temporal leakage|memory biases|risk.adjusted harm).{0,70}(?:financial|finance)|financial backtesting'), |
| ('financial_models','domain_models','Financial Pretraining & Adaptation / 金融预训练与领域适配',r'\b(?:FinBERT|FLANG|FLUE|ESGPT|Plutus)\b|financial.{0,65}(?:pretrain|pre-train|language model|LLMs?|domain adapt)|(?:pretrain|pre-train|domain adapt).{0,65}financial'), |
| ('financial_models','financial_benchmarks','Cross-Task Financial Evaluation / 跨任务金融综合评测',r'\b(?:FinBen|MultiFinBen|FinMME|CFMME|BizFinBench|VisFinEval|FinTrust)\b|financial.{0,40}(?:multimodal evaluation|application benchmark)|benchmarking.{0,50}(?:Greek Finance|finance domain)'), |
| ('credit_risk','credit_data','Lending & Credit Data / 借贷与信贷数据应用',r'\bloans?\b|\blending\b|mortgage|credit (?:graph|network|loan|agreement|line|card|balance)|\bcreditcard\b|default credit|consumer finance|assessment of credit|\bCredit[- ]Card\b|\bcredit8\b'), |
| ('fraud_aml','banknote_authentication','Banknote Authentication / 钞票真伪识别',r'banknote'), |
| ('banking','bank_data','Banking Data Applications / 银行业务数据应用',r'\bBank dataset\b|\bBank\b.{0,12}\d?\.\d{3}|bank accounts?|Bank of America'), |
| ('financial_nlp','nlp_datasets','Financial Language Tasks & Corpora / 金融语言任务与语料',r'financial NLP|financial domain RE|financial event types|Earnings-21|financial numeric entity|finance news|金融\(Finance\)|财务\(Finance\)'), |
| ('portfolio','investment_evaluation','Investment Evaluation & Decisions / 投资评价与决策',r'sequential investment|financial rebalancing|mean.field portfolio|mean.variance portfolio|optimal consumption and investment|investment decisions|low quality funds|portfolio (?:study|problem|game|risk management)|machine learning portfolio'), |
| ('financial_qa','qa_tasks','Financial Reasoning Tasks / 金融推理任务',r'financial QA|financial credit agreements|financial (?:and scientific|or scientific).{0,40}numeric|numeric.{0,45}financial|Finance Formula|TheoremQA|financial laws|finance reports'), |
| ('document_ai','financial_tables','Financial Tables & Spreadsheets / 财务表格与电子表格',r'FinTabNet|spreadsheets for finance|financial reports paired|tables.{0,90}finance|finance.{0,50}tables'), |
| ('time_series','financial_processes','Financial Stochastic Processes / 金融随机过程',r'stochastic volatility|financial data prediction|stock market models|\bNASDAQ series\b|stock market data|finance and weather forecasting|30 P ORTFOLIOS|Financial Trans.{0,5}actions'), |
| ('financial_models','numerical_finance','Numerical Finance / 计算金融与数值方法',r'financial integral|financial operations|finance on point clouds|financial.{0,20}PDE'), |
| ('financial_safety','financial_regulation','Financial Regulation / 金融监管与业务规则',r'financial regulations|financial laws|financial.{0,40}regulatory|FinLLM-Eval'), |
| ('financial_nlp','claim_verification','Financial Claim Verification / 金融声明核验',r'FinDVer|financial.{0,30}(?:claim verification|fact.verification)|(?:claim verification|fact.verification).{0,40}financ'), |
| ('portfolio','financial_planning','Financial Planning / 财务规划',r'financial planning|financial portfolios'), |
| ('corporate_finance','real_estate','Real Estate & Business Finance / 房地产与企业财务应用',r'real estate|real-estate|realty|housing.{0,30}(?:financial|investment)|financial properties of the housings'), |
| ('financial_models','financial_reasoning_eval','Professional Financial Evaluation / 金融专业能力评测',r'ProfBench|PRBench|Finance MBA|FinTMMBench'), |
| ('financial_safety','robustness_attacks','Adversarial Robustness in Finance / 金融应用中的攻击与防御',r'finance.{0,20}backdoor|backdoor.{0,30}finance'), |
| ('asset_pricing','risk_factors','Risk Factor Models / 市场风险因子建模',r'stock data factor analysis|financial factor model|lookahead factor model|factor models for quantitative investing'), |
| ('financial_nlp','financial_translation','Financial Translation / 金融翻译',r'translat.{0,60}(?:financial|finance|tax)|financial.{0,60}translat'), |
| ('banking','personal_finance','Personal Finance Interaction / 个人理财交互',r'personal finance|finance.{0,15}(?:budgets|budgeting)|financial advice'), |
| ] |
|
|
| |
| |
| |
| FALLBACK_RULES = [ |
| ('financial_safety','privacy_security',r'jailbreak|security|harmless|harmful|saniti|leakage|honeypot|extraction attacks|red.team|deception|ideolog|hijack|safety|refusal|policy alignment'), |
| ('financial_ir','retrieval',r'retriev|rerank|re.rank|information.seek|text matching|text embedding|query.focused'), |
| ('financial_qa','qa_tasks',r'question answer|\bQA\b|mathematical reasoning|number prediction|tabular reasoning|table understanding|aggregation queries|quantity extractor'), |
| ('financial_nlp','information_extraction',r'named entity|\bNER\b|entity recognition|relation extraction|information extraction|word segmentation|sequence labeling'), |
| ('financial_nlp','text_generation',r'summariz|summaris|text generation|writing|news comment generation|simplification|text.to.table|translating headers'), |
| ('financial_nlp','sentiment_stance',r'sentiment|stance detection'), |
| ('financial_nlp','nlp_datasets',r'text classifi|text categor|dialectal corpus|word sense|ChineseBERT|language translation|narrative media|news.{0,40}fake news'), |
| ('document_ai','financial_charts',r'visualization|visual semantics|visual.semantic|multimodal.{0,40}comprehension|General.Bench|vision.language'), |
| ('financial_agents','agent_decisions',r'\bagents?\b|negotiation|tool.call|tool.using|MCP|FutureX|ForecastBench|ForecastCompass|InsightBench|AutoDiscovery'), |
| ('banking','customer_intents',r'conversational|dialog|acoustic memory'), |
| ('trading','microstructure',r'clear the market|transaction costs'), |
| ('credit_risk','financial_risk',r'CVaR|risk.aware|model risk management'), |
| ('time_series','financial_processes',r'data streams|changepoint|score climbing|stochastic process|spatio.temporal'), |
| ('financial_models','domain_models',r'domain adapt|fine.tun|model merging|models.{0,20}merg|pretrain|pre.train|distillat|domain.specific|specialized generalist|memory decoder|AMARIS|VersaTune|REEF|LLM DNA|GEPA'), |
| ('financial_models','financial_benchmarks',r'benchmark|evaluation|evaluating|reasoning|LLMs?|language models?|test.time scaling'), |
| ] |
|
|
| METHODS = [ |
| ('llm','Language Models & LLM / 语言模型与LLM',r'\bLLMs?\b|language models?|instruction.tuning|GPT[- ]?\d|\b(?:Llama|Qwen|BERT)\b'), |
| ('reinforcement_learning','Reinforcement Learning / 强化学习',r'reinforcement learning|\bRL\b|bandit|GFlowNet|Q.learning'), |
| ('graph_learning','Graph Learning / 图学习',r'graph neural|graph learning|graph convolution|graph attention|\bGNNs?\b|hypergraph|graph fraud'), |
| ('diffusion_flow','Diffusion & Flow / 扩散与流模型',r'diffusion|flow matching|normalizing flows?|rectified flow'), |
| ('generative','Generative Modeling / 生成建模',r'generative|synthetic data|data synthesis|\bGANs?\b|variational autoencoder|data generation'), |
| ('transformer','Transformer / Transformer模型',r'transformer|self.attention'), |
| ('causal','Causal Learning / 因果学习',r'causal|structural causal|causality'), |
| ('fairness_recourse','Fairness & Recourse / 公平性与算法追索',r'fairness|fair learning|fair representation|counterfactual explanation|algorithmic recourse|strategic classification'), |
| ('uncertainty_robustness','Uncertainty & Robustness / 不确定性与鲁棒性',r'uncertainty|conformal|robust|distribution shift|adversarial|calibration'), |
| ('privacy','Privacy & Federated Learning / 隐私与联邦学习',r'differential privacy|differentially private|privacy.preserving|federated learning|private learning|unlearning'), |
| ('tabular','Tabular Learning / 表格学习',r'tabular|decision tree|random forest|gradient boost|TabPFN|TabM\b|TabDPT'), |
| ('multimodal','Multimodal Learning / 多模态学习',r'multi.modal|cross.modal|vision.language|visual.language|audio.text|text.audio'), |
| ('retrieval','Retrieval & RAG / 检索与RAG',r'\bRAG\b|retrieval|re-rank|rerank|search engine'), |
| ('optimization','Optimization & Numerical Methods / 优化与数值方法',r'optimization|optimisation|optimal control|differential equations?|convex|gradient descent|subgradient|PDE|branch.and.bound'), |
| ('representation','Representation Learning / 表征学习',r'representation learning|contrastive learning|embeddings?|pre.training|pretraining'), |
| ] |
|
|
| def compact(text): |
| return re.sub(r'\s+',' ',re.sub(r'[\x00-\x08\x0b-\x1f]',' ',str(text or ''))).strip() |
|
|
| def classify(paper, info, override=None): |
| sources=[('title',compact(paper['title']),100), |
| ('finance_evidence',compact(info.get('evidence_excerpt')),65), |
| ('abstract',compact(paper.get('abstract')),40)] |
| evidence=[];scores={};subtopics=[] |
| for parent,ident,label,pattern in RULES: |
| matches=[] |
| for source,text,score in sources: |
| match=re.search(pattern,text,re.I) |
| if match: |
| matches.append(dict(topic_id=parent,subtopic_id=ident,label=label,source=source, |
| matched=match.group(),excerpt=text[max(0,match.start()-100):match.end()+180],score=score)) |
| if matches: |
| best=max(matches,key=lambda x:x['score']) |
| evidence.append(best);subtopics.append(ident) |
| scores[parent]=max(scores.get(parent,0),best['score']) |
| if not scores: |
| for parent,ident,pattern in FALLBACK_RULES: |
| m=re.search(pattern,sources[0][1],re.I) |
| if m: |
| label=next(r[2] for r in RULES if r[1]==ident) |
| scores[parent]=25;subtopics.append(ident) |
| evidence.append(dict(topic_id=parent,subtopic_id=ident,label=label, |
| source='title_and_confirmed_finance_evidence',matched=m.group(), |
| excerpt='Title: '+sources[0][1]+' | Financial application: '+sources[1][1][:350],score=25)) |
| break |
| if not scores: |
| scores['other_finance']=0 |
| evidence.append(dict(topic_id='other_finance',subtopic_id=None,source='finance_evidence', |
| matched='confirmed finance inclusion',excerpt=compact(info.get('evidence_excerpt'))[:350],score=0)) |
| |
| |
| priority=['alpha_mining','derivatives','asset_pricing','portfolio','trading','credit_risk','fraud_aml','insurance', |
| 'market_prediction','financial_qa','financial_ir','document_ai','esg','macro_economics', |
| 'corporate_finance','banking','financial_safety','financial_agents','time_series','blockchain', |
| 'financial_nlp','financial_models','other_finance'] |
| primary=max(scores,key=lambda k:(scores[k],-priority.index(k))) |
| topic_ids=[x[0] for x in TOPICS if x[0] in scores] |
| methods=[];method_evidence=[] |
| for ident,label,pattern in METHODS: |
| for source,text,_ in [sources[0],sources[2]]: |
| m=re.search(pattern,text,re.I) |
| if m: |
| methods.append(ident);method_evidence.append(dict(method_id=ident,source=source,matched=m.group(), |
| excerpt=text[max(0,m.start()-65):m.end()+100]));break |
| result=dict(primary_topic_id=primary,topic_ids=topic_ids,subtopic_ids=subtopics,method_ids=methods, |
| topic_evidence=evidence,method_evidence=method_evidence) |
| if override: |
| for key in ['primary_topic_id','topic_ids','subtopic_ids','method_ids']: |
| if key in override:result[key]=override[key] |
| assert result['primary_topic_id'] in result['topic_ids'] |
| if 'topic_ids' in override and 'subtopic_ids' not in override: |
| parents={r[1]:r[0] for r in RULES} |
| result['subtopic_ids']=[s for s in result['subtopic_ids'] if parents[s] in result['topic_ids']] |
| result['topic_evidence']=[e for e in result['topic_evidence'] if e['topic_id'] in result['topic_ids']] |
| if override.get('evidence_excerpt'): |
| for topic_id in result['topic_ids']: |
| if not any(e['topic_id']==topic_id for e in result['topic_evidence']): |
| result['topic_evidence'].append(dict(topic_id=topic_id,subtopic_id=None, |
| source='manual_review',matched=override['reason'],excerpt=override['evidence_excerpt'],score=110)) |
| result['manual_override']=override |
| |
| ts_ids={i for parent,i,label,rx in RULES if parent=='time_series'} |
| if len(ts_ids.intersection(result['subtopic_ids']))>1: |
| result['subtopic_ids']=[s for s in result['subtopic_ids'] if s!='ts_general'] |
| return result |
|
|
