Extracting Personal Names from Applying Named Entity Recognition to Informal Text Einat Minkov & Richard C. Wang Language Technologies Institute William W. Cohen Center for Automated Learning and Discovery School of Computer Science Carnegie Mellon University
October 7, 2005CMU School of Computer Science2 What is an informal text? A text that is… –Written for a narrow audience Group/task-specific abbreviations often used Not self-contained (context shared by a related group of people) –Not carefully prepared Contains grammatical and spelling errors Does not follow capitalization conventions Some examples are… –Instant messages –Newsgroup postings – messages
October 7, 2005CMU School of Computer Science3 Objective / Outline Investigate named entity recognition (NER) for informal text –Conduct experiments on recognizing personal names in Examine indicative features in and newswire Suggest specialized features for Evaluate performance of a state-of-the-art extractor (CRF) Analyze repetition of names in and newswire Suggest and evaluate a recall-enhancing method that is effective for
October 7, 2005CMU School of Computer Science4 Corpora Mgmt corpora – s from a management course at CMU in which students form teams to run simulated companies –Teams: Each set (train/tune/test) formed by different simulation teams –Game: Each set formed by different days during the simulation period Enron corpora – s from Enron Corporation –Meetings: Each set formed by randomly selected meeting-related s –Random: Each set formed by repeatedly sampling a user then sampling an from that user, both at random Note: The number of words and names refer to the whole annotated corpora
October 7, 2005CMU School of Computer Science5 Extraction Method Train Conditional Random Fields (CRF) to label and extract personal names –A machine-learning based probabilistic approach to labeling sequences of examples Learning reduces NER to the task of tagging, or classifying, each word using a set of five tags: –Unique: A one-token entity –Begin: The first token of a multi-token entity –End: The last token of a multi-token entity –Inside: Any other token of a multi-token entity –Outside: A token that is not part of an entity Example: Einat and Richard Wang met William W. Cohen today Unique Outside Begin End Outside Begin Inside End Outside
October 7, 2005CMU School of Computer Science6 Top Learned Features Features most indicative of a token being part of a name in a Conditional Random Fields (CRF) extractor Note: A feature is denoted by its direction (left/right) comparing to the focus word, offset, and lexical value Newswire (MUC-6) (Mgmt-Game) 2 In Quoted Excerpt In Signature Name Titles Job Titles Results show that… and newswire text have very different characteristics
October 7, 2005CMU School of Computer Science7 Note: All features are instantiated for the focus word t, and 3 tokens to the left and right of t Our Proposed Features
October 7, 2005CMU School of Computer Science8 Feature Evaluation Entity-level F1 of learned extractor (CRF) using: –Basic features (B) –Basic and features (B+E) –Basic and Dictionary features (B+D) –All features (B+D+E) B+D+E PrecisionRecall Results show that… 1) Dictionary and features are useful (best when combined) 2) Generally high precision but low recall
October 7, 2005CMU School of Computer Science9 What’s Next? Previous experiments show high precision but low recall –Next goal: Improve recall One recall-enhancing method –Look for multiple occurrences of names in a corpus We conduct experimental studies –Examine repetition patterns of names in and newswire text –Examine occurrences of names within a single document and across multiple documents
October 7, 2005CMU School of Computer Science10 Doc. Frequency of Names Percentage of person-name tokens that appear in at most K distinct documents as a function of K 1 Document Frequency Percentage 30% of names in Mgmt-Game appear only in one document Nearly 80% of names in MUC-6 appear only in one document About 20% of names in Mgmt- Game appear in 10+ documents Only 1.3% of names in MUC-6 appear in 10+ documents Results show that… Repetition of names across multiple documents is more common in corpora unique(A) : duplicates removed from set A df(w) : # of documents containing token w
October 7, 2005CMU School of Computer Science11 Single vs. Multiple Documents We define the following extractors: 1.CRF – baseline trained with all features 2.SDR (Single Document Repetition) Rules that extract person-name tokens that appear more than once within a single document; hence an upper bound on recall using only names repetition within a single document 3.MDR (Multiple Document Repetition) Rules that extract person-name tokens that appear in more than one document; hence an upper bound on recall using only names repetition across multiple documents 4.SDR+CRF Union of extractions by SDR and CRF; hence an upper bound on recall using CRF and names repetition within a single document 5.MDR+CRF Union of extractions by MDR and CRF; hence an upper bound on recall using CRF and names repetition across multiple documents
October 7, 2005CMU School of Computer Science12 Single vs. Multiple Documents Token-level upper bounds on recall and potential recall-gains associated with methods that look for name tokens that re-occur within a single document or across multiple documents Results show that… Higher recall and potential recall-gains can be obtained for corpora using MDR method MUC-6 has highest recall-gain using SDR MUC-6 has highest recall using SDR MUC-6 has lowest recall using MDR MUC-6 has lowest recall-gain using MDR
October 7, 2005CMU School of Computer Science13 What’s Next? Our studies show the potential of exploiting repetition of names over multiple documents for improving recall in corpora We suggest a recall-enhancing method: 1.Auto-construct a dictionary of predicted names and their variants from test set 2.Statistically filter out noisy names from the dictionary 3.Match names globally from the inferred dictionary onto test set, exploiting repetition of names Note: A “dictionary” is simply a list of one or more tokens
October 7, 2005CMU School of Computer Science14 Name Dictionary Construction Every name in the test set predicted by the learned extractor (CRF), trained with all features, is transformed into a set of name variants and inserted into a dictionary Transformation Example Name variants of “Benjamin Brown Smith”. Original name is included by default
October 7, 2005CMU School of Computer Science15 Name Dictionary Filtering Previously constructed dictionary contains noisy names –i.e. “brown” can also refer to a color –Next goal: Filter out noisy names We suggest a filtering scheme to remove every single- token name w from the dictionary when PF.IDF(w) < Θ cpf(w) : # of times w is predicted as a name-token in corpus ctf(w) : # of occurrences of w in corpus df(w) : document frequency of w in corpus N : # of documents in corpus Words that get low PF.IDF scores are either highly ambiguous names or very common words in corpus Note: “Corpus” mentioned here refers to the test set in our experiments Θ = 0.16 optimizes entity- level F1 in tune sets; thus, we apply the same threshold onto our test sets Predicted Frequency × Inverse Document Frequency
October 7, 2005CMU School of Computer Science16 Name Matching I called Benjamin Brown Smith and left a message to send us an if he could come. I have not received his yet. He might not be able to come. We may want to postpone until tomorrow morning. Do you still have our class schedule? Please contact benjamin and confirm the meeting. I do not have classes tomorrow morning. A window slides through every token in the test set A match occurs when tokens in a window starts with the longest possible name variant in the dictionary All matched names are marked for evaluation … benjamin brown smith benjamin-brown smith benjamin brown-smith benjamin-brown-smith benjamin brown s. benjamin-b. smith benjamin b. smith benjamin brown-s. benjamin-brown s. benjamin-brown-s benjamin-b. s. benjamin-smith benjamin smith b. brown smith benjamin b. s. b. brown-smith benjamin-s. benjamin s. b. brown s. b. b. smith b. brown-s. benjamin b. smith b. b. s. smith b. s. … Filtered Dictionary Names Matching Example Predicte d by CRF Missed by CRF
October 7, 2005CMU School of Computer Science17 Experimental Results Entity-level relative improvements (and final scores) after applying our recall-enhancing method on test sets –Baseline: learned extractor (CRF) trained with all features Results show that… 1) Recall improved significantly with small sacrifice in precision 2) F1 scores improved in all cases
October 7, 2005CMU School of Computer Science18 Conclusion and newswire text have different characteristics We suggested a set of specialized features for names extraction on exploiting structural regularities in Exploiting names repetition over multiple documents is important for improving recall in corpora We presented the PF.IDF recall-enhancing method that improves recall significantly with small sacrifice in precision
October 7, 2005CMU School of Computer Science19 Thank You!
October 7, 2005CMU School of Computer Science20 References