Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Giveme5W1H: A Universal System for Extracting Main Events from News Articles

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An open-source system extracts a news article's main event by answering the journalistic 5W1H questions, achieving 0.73 mean average generalized precision overall and 0.82 on the first four Ws.

desk verdict The open-source system and dataset are the real contribution; the 0.73/0.82 evaluation numbers should be treated as upper-bound estimates until a same-protocol baseline or independent gold standard appears. read the letter →

arxiv 1909.02766 v1 pith:3K6ZYXBA submitted 2019-09-06 cs.CL

Felix Hamborg , Corinna Breitinger , Bela Gipp This is my paper · ORCID
classification cs.CL
keywords eventextraction5W1Hnewsanalysisquestionansweringmaindetectioncanonicalizationopen-sourcesoftwarenaturallanguageprocessing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Giveme5W1H is an open-source system that takes an English news article and returns short phrases answering who did what, when, where, why, and how; together these six answers describe the article's main event. The paper's central claim is that this universal, rule-based approach is accurate enough for real use: in an expert evaluation on 120 news articles it achieved a mean average generalized precision of 0.73 over all six questions and 0.82 over the first four Ws, which the authors argue can alone summarize an event. This matters because event extraction is a prerequisite for news aggregation, clustering, summarization, and media-bias analysis, and no equally general, publicly available tool existed. The paper also contributes an annotated training dataset and a modular pipeline, letting other researchers reuse, adapt, and benchmark against the system instead of reimplementing extraction from scratch.

What carries the argument

The central machinery is the 5W1H question set used as an event schema, operationalized by four extraction chains. The action chain collects subject noun phrases and their sibling verb phrases for who and what; the environment chain uses canonicalized temporal and geographic candidates for when and where; the cause chain detects causal conjunctions, causative adverbs, and causative-verb patterns for why; and the method chain finds copulative-conjunction clauses and adjective/adverb phrases for how. Each candidate is scored by a weighted sum $s_q = \sum_i w_{q,i} s_{q,i}$ of position, frequency, and type-specific factors, with weights tuned by a semi-automated search on the annotated dataset using semantic and temporal/geographic error distances; a combined scorer then adjusts method scores by the sentence distance to the chosen action. This combination of canonicalization, parse-tree rules, and learned weights carries the extraction from raw article text to a concise event description.

What would settle it

Take a held-out set of more than 100 articles from outlets and beats not represented in the training data, such as local newspapers or financial wire services, have three annotators score the extracted phrases on the same graded relevance scale, and check whether mean average generalized precision on the first four W questions remains near 0.82; if it falls well below that, the tuned configuration is tied to the training distribution rather than being universal.

Watch

Extended reading notes

Core claim

The paper's central claim is that the journalistic 5W1H questions are a workable operational definition of a news article's main event, and that a carefully engineered pipeline of syntactic and domain-specific rules can answer them from explicit text with usable precision. The system canonicalizes temporal mentions to standardized timestamps, locations to geocoordinates, and other named entities to knowledge-graph concepts; extracts candidates through four dedicated chains; and selects final answers with per-question weighted scoring plus a combined sentence-distance adjustment, with roughly twenty scoring weights learned semi-automatically from a new annotated dataset. In the reported evaluation, mean average generalized precision is strong for who (0.92), what (0.79), when (0.78), and where (0.78), but weaker for why (0.48) and how (0.61). The authors further state that this is the only universal, open-source 5W1H extractor, and that it outperforms their earlier five-W system.

Load-bearing premise

The reported accuracy assumes that the roughly twenty scoring weights tuned on an 80-article training sample from 13 major US and UK outlets will perform just as well on news articles generally, including other outlets, genres, and writing styles.

Editorial extensions

If this is right

  • Researchers can use the system and its datasets directly, avoiding the redundant reimplementation of event extraction that motivated the paper.
  • Downstream tasks such as article clustering, summarization, and news aggregation can consume six standardized phrases, with temporal and geospatial information in canonical form.
  • Because the first four Ws alone reach 0.82 precision, applications that need only who, what, when, and where can treat that subset as sufficiently reliable for many analyses.
  • The modular architecture allows future work to replace individual components, such as the cause or method extractors, without discarding the rest of the pipeline.
  • The newly released annotated dataset gives the field a shared training and evaluation resource, making future system comparisons possible for the first time.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: the scoring weights tuned on roughly 80 training articles from major US and UK outlets may not transfer to local, wire-service, or non-Western news styles; a cross-outlet generalization test would reveal how universal the 'universal' claim really is.
  • Inference: the weaker why and how scores suggest the bottleneck is not the 5W1H schema itself but the detection of explicitly signaled causes and methods; a learned or hybrid component for those two questions could be slotted into the modular pipeline and tested against the same benchmark.
  • Inference: if the 0.82 for the first four Ws holds across domains, main-event extraction for explicitly reported events becomes a nearly solved subtask, shifting research attention to implicit events, cross-document event coreference, and canonical event representation.
  • Inference: the reported numbers are not directly comparable to earlier precision figures because relevance scales and test sets differ; a shared evaluation corpus would let the community arbitrate between systems.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents Giveme5W1H, an open-source system that extracts phrases answering the journalistic 5W1H questions (who, what, when, where, why, how) from English news articles to describe the main event. It describes a three-phase pipeline (preprocessing with CoreNLP, SUTime, Nominatim, and AIDA; four extraction chains for action, environment, cause, and method; and scoring functions with learned weights), and reports an expert evaluation on 120 BBC articles with three assessors, giving a mean average generalized precision (MAgP) of 0.73 across all questions and 0.82 for the first four W questions. The paper also introduces an annotated training dataset and states that the system is the only universal open-source 5W1H extractor.

Significance. If the reported evaluation is credible, this paper fills a genuine gap: it provides a modular, publicly available system and an annotated dataset, plus a sufficiently detailed algorithm description to support reimplementation and benchmarking. The release of the code and the dataset is a concrete contribution that should be credited. The headline numbers (MAgP 0.73/0.82) would demonstrate useful performance for news analysis use cases, provided the evaluation methodology is sound. The paper is clearly written in most technical sections, and the modular design is a strength. However, the evaluation protocol has weaknesses that directly affect the central claim, so the findings should be treated as promising but not yet fully substantiated.

major comments (3)
  1. [Section 4, Table 6] The evaluation uses a display-and-rate protocol in which assessors see the phrases Giveme5W1H has extracted and judge them as relevant, partially relevant, or non-relevant, without an independent gold-standard annotation of the test articles and without a baseline system evaluated under the same protocol. Because Section 5 states that the system always outputs a 'why' or 'how' candidate when one is found, raters may anchor on plausible but non-central phrases and award partial credit, which can inflate MAgP. This is a direct threat to the central numerical claim of 0.73/0.82. The paper should either (a) provide an independent gold-standard annotation of the 120 test articles (or a subset) and report agreement between system output and that gold standard, or (b) run at least one baseline (e.g., Giveme5W, or a simple lead/headline heuristic) under the identical rating protocol and report the difference with a significance test. Reporting only inter-assessor agreement does not address the anchoring concern, since the raters are rating the same displayed phrases.
  2. [Section 3.5] The scoring weights and heuristic thresholds (on the order of twenty parameters) are tuned on an 80-article training set using a semi-automated search, with only 20 articles held out for parameter validation and no cross-validation or stability analysis. The evaluation set of 120 BBC articles is separate, which is good, but the paper provides no variance estimates, confidence intervals, or significance tests for the reported MAgP values. The claim that the configuration generalizes from the training sample to news articles at large is therefore unquantified. The authors should report bootstrap or split-half variability of the MAgP on the 120 articles, and, if possible, show how MAgP varies when the parameter search is repeated on different training subsamples. Without this, the reader cannot assess how much of the reported performance is due to overfitting to the annotation scheme or to the specific training outlets.
  3. [Section 5, 'why'/'how' output] The paper acknowledges that for 'why' and 'how' the system always outputs a candidate as long as one is found, with no score threshold to suppress low-confidence answers. The 'why' MAgP is 0.48, which is well below the other questions. This conflation of ranking quality with the absence of abstention makes the practical utility of the output unclear: a user applying the system would receive 'why' and 'how' phrases that are frequently non-relevant, yet the reported aggregate MAgP treats them as part of the overall performance. The authors should provide a threshold-abstention analysis (e.g., MAgP as a function of a score cutoff, or precision/recall with an abstain option) to clarify whether the current always-return behavior is appropriate for the claimed use cases, or whether a threshold should be set in the released system.
minor comments (5)
  1. [Section 3.5] The phrase 'ICR/44=0.81' appears to contain a typo or LaTeX artifact; this should read 'ICR = 0.81'.
  2. [Section 3] The workflow description says 'as shown in Figure 1', but Figure 1 is the news article example; the pipeline diagram is Figure 2, so the reference should be corrected.
  3. [Section 4, Table 6] The table shows ICR values in the first numeric column without a caption note explaining this; separating the ICR row from the per-category columns would improve readability.
  4. [Section 3.3] The formula for the weighted score sum is garbled in the rendering: 's;=∑wq,?sq,?@A"?BC' should be typeset properly, e.g., s_q = sum_i w_{q,i} s_{q,i}.
  5. [Section 4] The comparison with Parton et al. states 'Giveme5W1H achieves a 0.12 higher MAgP}' with a malformed superscript; this should read MAgP_5W.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: reported MAgP is an independent held-out evaluation, not a reconstruction of its inputs.

full rationale

The paper's central claim is the measured MAgP of 0.73 (all questions) and 0.82 (first four W questions) on 120 BBC test articles. The derivation chain is empirical rather than definitional: the system extracts candidate phrases using syntactic and domain-specific rules (Section 3.2), scores candidates with weighted factors whose values are selected on a separate annotated training set (Section 3.5), and is then evaluated on held-out BBC articles by three assessors who read each article and rate the relevance of the system's output phrases on a 0/0.5/1 scale (Section 4). The evaluation set is distinct from the training set used to fit the scoring weights, so the headline numbers are not fitted inputs renamed as predictions. The paper's citations to the authors' prior Giveme5W work serve as a baseline comparison and as background for the claim that no universal open-source 5W1H system existed; those citations are contextual and do not by themselves force the reported performance numbers. The evaluation protocol may be vulnerable to rater anchoring because the assessors view the system's own outputs, but that is a validity or correctness concern, not a circularity: no equation or definition in the paper reduces the reported MAgP to the training annotations or to a self-citation. The discussion in Section 5 about the absence of a score threshold for low-confidence 'why' and 'how' answers is an acknowledged limitation, not a circular step. No specific reduction of a claimed result to its inputs can be exhibited, so under the hard rules the appropriate finding is no significant circularity.

Assumptions & free parameters 30 free parameters · 8 assumptions · 0 invented entities

The central claim depends on roughly twenty scoring weights and cutoff constants tuned on an 80-article training set, plus several domain assumptions about news writing (inverted pyramid, frequent mention of main actors, temporal locality of events). No new entities are postulated. The evaluation itself adds a test-set assumption: the 120 BBC articles and assessor judgments are representative.

free parameters (30)
  • who_position_weight = 0.9
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • who_frequency_weight = 0.095
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • who_type_weight = 0.005
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • when_position_weight = 0.24
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • when_frequency_weight = 0.16
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • when_closeness_weight = 0.4
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • when_duration_weight = 0.2
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • where_position_weight = 0.37
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • where_frequency_weight = 0.3
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • where_containment_weight = 0.3
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • where_specificity_weight = 0.03
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • why_position_weight = 0.56
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • why_type_weight = 0.44
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • how_position_weight = 0.23
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • how_frequency_weight = 0.14
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • how_type_weight = 0.63
    Learned via semi-automated search on 80 training articles (Section 3.5).
  • causal_conjunction_reliability = 1.0
    Reliability constant for causal conjunctions in 'why' scoring (Section 3.3).
  • causative_adverb_reliability = 0.62
    Reliability constant for causative adverbs in 'why' scoring (Section 3.3).
  • causative_verb_reliability = 0.06
    Reliability constant for causative verbs in 'why' scoring (Section 3.3).
  • copulative_conjunction_reliability = 1.0
    Reliability constant for copulative conjunction 'how' candidates (Section 3.3).
  • adjective_adverb_fallback_reliability = 0.41
    Reliability constant for adjective/adverb fallback 'how' candidates (Section 3.3).
  • locality_merge_range = 1
    Hand-set range for merging locality named entity tokens (Section 3.1).
  • what_min_tokens = 3
    Hand-set cutoff to avoid undescriptive 'what' phrases (Section 3.2).
  • how_max_tokens = 10
    Hand-set cutoff for 'how' phrase length (Section 3.2).
  • combined_distance_weight = 1
    Weight for the combined sentence-distance scorer in Eq. 1 (Section 3.3).
  • one_month_seconds = 2.5e6
    Normalization bound for temporal closeness scoring (Section 3.3).
  • min_duration_seconds = 60
    Lower bound for temporal duration normalization (Section 3.3).
  • max_duration_seconds = 3.1e7
    Upper bound for temporal duration normalization (Section 3.3).
  • min_area_sqm = 225
    Lower bound for location specificity normalization (Section 3.3).
  • max_area_sqm = 5.3e11
    Upper bound for location specificity normalization (Section 3.3).
assumptions (8)
  • domain assumption The 5W1H questions adequately describe a news article's main event.
    Stated in the introduction and abstract; the entire extraction target depends on this.
  • domain assumption Events occur at a single point in time with a short duration.
    Used to filter temporal candidates in preprocessing (Section 3.1).
  • domain assumption Main event information is concentrated early in the article.
    Motivates the position scoring factor for who, when, where, why, and how (Section 3.3).
  • domain assumption Main actors are mentioned frequently in an article.
    Motivates the frequency scoring factor (Section 3.3).
  • domain assumption A method phrase tends to appear close to the action phrase.
    Basis for Eq. 1, the combined sentence-distance scorer (Section 3.3).
  • domain assumption Word Mover's Distance is a valid semantic similarity measure for phrase-level annotation matching.
    Used as the error function for parameter learning on who, what, why, and how (Section 3.5).
  • domain assumption The 100-article annotated dataset and the 120-article BBC evaluation set are representative of English news.
    Parameter tuning and evaluation rest on these samples (Sections 3.5 and 4).
  • domain assumption Assessor judgments on the 3-point relevance scale accurately reflect answer quality.
    The MAgP scores are derived from these judgments (Section 4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Giveme5W1H: A Universal System for Extracting Main Events from News Articles." pith.science (2026). https://pith.science/paper/3K6ZYXBA

@misc{pith2026190902766,
  author       = {Pith},
  title        = {Pith review of: Giveme5W1H: A Universal System for Extracting Main Events from News Articles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3K6ZYXBA}},
  note         = {Machine review of arXiv:1909.02766}
}
read the original abstract

Event extraction from news articles is a commonly required prerequisite for various tasks, such as article summarization, article clustering, and news aggregation. Due to the lack of universally applicable and publicly available methods tailored to news datasets, many researchers redundantly implement event extraction methods for their own projects. The journalistic 5W1H questions are capable of describing the main event of an article, i.e., by answering who did what, when, where, why, and how. We provide an in-depth description of an improved version of Giveme5W1H, a system that uses syntactic and domain-specific rules to automatically extract the relevant phrases from English news articles to provide answers to these 5W1H questions. Given the answers to these questions, the system determines an article's main event. In an expert evaluation with three assessors and 120 articles, we determined an overall precision of p=0.73, and p=0.82 for answering the first four W questions, which alone can sufficiently summarize the main event reported on in a news article. We recently made our system publicly available, and it remains the only universal open-source 5W1H extractor capable of being applied to a wide range of use cases in news analysis.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Automated Journalistic Questions: A New Method for Extracting 5W1H in French

    cs.CL 2025-05 conditional novelty 6.0 of 10

    The first French 5W1H extraction pipeline matches GPT-4o in agreement with human annotators on a new 250-article Quebec news corpus.

Reference graph

Works this paper leans on

49 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    He drove quickly

    Thus far, no systems have been described in sufficient detail to allow for a reimplementation by other researchers. Both the ‘why’ and ‘how’ question pose a particular challenge in comparison to the other questions. As discussed by Hamborg et al. [17], determining the reason or cause (i.e. ‘why’) can even be difficult for humans. Often the reason is unkno...

    work page 2009

  2. [5]

    We determine the number of mentions of a method phrase 𝑛a(𝑐) by the term frequency (including in-flected forms) of its most frequent token (cf

    Table 5: Weights and scoring factors for ‘how’ phrases 𝑖 𝑤how,? 𝑠how,? 0 (position) .23 pos(𝑐) 1 (frequency) .14 f(𝑐) 2 (type) .63 TM(𝑐) The method type TM(𝑐)=1 if 𝑐 is extracted because of a copula-tive conjunction, else 0.41. We determine the number of mentions of a method phrase 𝑛a(𝑐) by the term frequency (including in-flected forms) of its most frequ...

    work page 2019

  3. [6]

    and can be used by other re-searchers to train their own 5W1H approaches. This paper is rel-evant to researchers and developers from various disciplines with the shared aim of extracting and analyzing the main events that are being reported on in articles. INRA 2019, September 2019, Copenhagen, Denmark Hamborg et al. 2 Figure 1: News article

    work page 2019

  4. [12]

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2012), 540–555

    The 5W structure for sentiment summarization-visualization-tracking. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2012), 540–555

    work page 2012

  5. [16]

    Proceedings of the ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL) (Urbana-Champaign, IL, USA, 2019), 1–10

    Automated Identification of Media Bias by Word Choice and Labeling in News Articles. Proceedings of the ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL) (Urbana-Champaign, IL, USA, 2019), 1–10

    work page 2019

  6. [18]

    International Journal on Digital Libraries

    Bias-aware news analysis using matrix-based news aggregation. International Journal on Digital Libraries. (2018). DOI:https://doi.org/10.1007/s00799-018-0239-9

    work page doi:10.1007/s00799-018-0239-9 2018

  7. [19]

    Proceedings of the ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL) (Fort Worth, Texas, USA, 2018), 339–340

    Extraction of Main Event Descriptors from News Articles by Answering the Journalistic Five W and One H Questions. Proceedings of the ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL) (Fort Worth, Texas, USA, 2018), 339–340

    work page 2018

  8. [20]

    widespread corruption in the finance ministry has cost it $2m

    – surprisingly almost identical to the ICR of our assessors. We found that different forms of journalistic presentation in the five news categories of the dataset led to different extraction performance. Politics articles, which yielded the best perfor-mance, mostly reported on single events. The performance on sports articles was unexpectedly high, even ...

    work page 2019

Show all 49 references
  1. [21]

    Proceedings of the iConference 2018 (Sheffield, UK, 2018), 355–356

    Giveme5W: Main Event Retrieval from News Articles by Extraction of the Five Journalistic W Questions. Proceedings of the iConference 2018 (Sheffield, UK, 2018), 355–356

    2018

  2. [23]

    Journal of the American Medical Informatics Association

    Agreement, the F-measure, and reliability in information retrieval. Journal of the American Medical Informatics Association. 12, 3 (2005), 296–298. DOI:https://doi.org/10.1197/jamia.M1733

    2005 doi

  3. [24]

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2005), 510–515

    TrackThem: Exploring a large-scale news video archive by tracking human relations. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2005), 510–515

    2005

  4. [25]

    Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics-Volume 1 (1998), 571–577

    Information Classification and Navigation Based on 5W1 H of the Target Information. Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics-Volume 1 (1998), 571–577

    1998

  5. [28]

    Literary and Linguistic Computing

    Automatic extraction of cause-effect information from newspaper text without knowledge-based inferencing. Literary and Linguistic Computing. 13, 4 (1998), 177–186

    1998

  6. [29]

    Proceedings of The 32nd International Conference on Machine Learning

    From Word Embeddings To Document Distances. Proceedings of The 32nd International Conference on Machine Learning. 37, (2015), 957–966

    2015

  7. [30]

    ((NP) Mr. Trump, ((VP) who stormed to a shock election victory on Wednesday)), ((VP) said it was […])

    using AIDA [19]. The YAGO graph is a state-of-the-art knowledge base, where nodes in the graph represent semantic concepts that are connected to other nodes through attributes and relations. The data is derived from other well-established knowledge bases, such as Wikipedia, Wo...

    2019

  8. [31]

    after [the train came off the tracks]

    and compare the verb and its syn-onyms with the list of causative verbs from Girju [11], which we also extended by their synonyms (cf. [11]). If there is at least one match, we take the last NP of the causative pattern as the ‘why’ candidate. To reduce false positives, we chec...

    2019

  9. [32]

    Artificial Intelligence in Medicine

    Multilingual event extraction for epidemic detection. Artificial Intelligence in Medicine. (2015). DOI:https://doi.org/10.1016/j.artmed.2015.06.005

    2015 doi

  10. [33]

    Proceedings of the {HLT-NAACL} 2003 Workshop on Analysis of Geographic References

    InfoXtract location normalization: a hybrid approach to geographic references in information extraction. Proceedings of the {HLT-NAACL} 2003 Workshop on Analysis of Geographic References. 1, (2003), 39–44. DOI:https://doi.org/10.3115/1119394.1119400

    2003

  11. [34]

    Proceedings of CIDR

    YAGO3: A Knowledge Base from Multilingual Wikipedias. Proceedings of CIDR. (2015), 1–11. DOI:https://doi.org/10.1016/j.jbi.2013.09.007

    2015 doi

  12. [36]

    We also investigated the performance of systems that are only capable of extracting 5W phrases

    or only evaluated the extracted ‘who’ and ‘what’ phrases of Japanese news articles [22]. We also investigated the performance of systems that are only capable of extracting 5W phrases. Our system achieves a MAgP}~=0.75, which is 0.05 higher than the MAgP of Giveme5W [17]. We a...

    2019

  13. [37]

    Journal of Case Studies

    Improving the Hook in Case Writing. Journal of Case Studies. 30, (2012), 1–6

    2012

  14. [38]

    Croatian medical journal

    Kipling’s guide to writing a scientific paper. Croatian medical journal. 43, 3 (2002), 262–7

    2002

  15. [39]

    Language Resources and Evaluation

    Multilingual and cross-domain temporal tagging. Language Resources and Evaluation. 47, 2 (2013), 269–298. DOI:https://doi.org/10.1007/s10579-012-9179-y

    2013 doi

  16. [41]

    Journal of Peace Research

    Introducing the UCDP Georeferenced Event Dataset. Journal of Peace Research. 50, 4 (2013), 523–532. DOI:https://doi.org/10.1177/0022343313484347

    2013 doi

  17. [45]

    Information Processing (ISIP), 2010 Third International Symposium on (2010), 484–489

    Chinese news event 5w1h elements extraction using semantic role labeling. Information Processing (ISIP), 2010 Third International Symposium on (2010), 484–489

    2010

  18. [46]

    Proceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD 08 (2008),

    A unified approach for schema matching, coreference and canonicalization. Proceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD 08 (2008),

    2008

  19. [47]

    INTERSPEECH (2009), 2703–2706

    Classification-based strategies for combining multiple 5-w question answering systems. INTERSPEECH (2009), 2703–2706

    2009

  20. [48]

    INTERSPEECH (2009), 2707–2710

    Combining semantic and syntactic information sources for 5-w question answering. INTERSPEECH (2009), 2707–2710

    2009

  21. [49]

    Proceedings of the 21st annual international ACM SIGIR conference on Research and development in information retrieval - SIGIR ’98 (1998), 28–36

    A study on retrospective and on-line event detection. Proceedings of the 21st annual international ACM SIGIR conference on Research and development in information retrieval - SIGIR ’98 (1998), 28–36

    1998

  22. [1992]

    Proceedings of the 4th conference on Message understanding (1992), 3–21

    Overview of the fourth message understanding evaluation and conference. Proceedings of the 4th conference on Message understanding (1992), 3–21

    1992

  23. [1995]

    Communications of the ACM

    WordNet: a lexical database for English. Communications of the ACM. 38, 11 (1995), 39–41. DOI:https://doi.org/10.1145/219717.219748

    1995

  24. [1998]

    Proceedings of the DARPA Broadcast News Transcription and Understanding Workshop (1998), 194–218

    Topic detection and tracking pilot study: Final report. Proceedings of the DARPA Broadcast News Transcription and Understanding Workshop (1998), 194–218

    1998

  25. [2000]

    American Journal of Sociology

    Political Processes and Local Newspaper Coverage of Protest Events: From Selection Bias to Triadic Interactions. American Journal of Sociology. 106, 2 (2000), 463–505

    2000

  26. [2002]

    Journal of the American Society for Information Science and Technology

    Using graded relevance assessments in IR evaluation. Journal of the American Society for Information Science and Technology. 53, 13 (2002), 1120–1129

    2002

  27. [2003]

    Proceedings of the ACL 2003 workshop on Multilingual summarization and question answering-Volume 12 (2003), 76–83

    Automatic detection of causal relations for question answering. Proceedings of the ACL 2003 workshop on Multilingual summarization and question answering-Volume 12 (2003), 76–83

    2003

  28. [2005]

    Proceedings of the 43rd annual meeting on association for computational linguistics (2005), 363–370

    Incorporating non-local information into information extraction systems by gibbs sampling. Proceedings of the 43rd annual meeting on association for computational linguistics (2005), 363–370

    2005

  29. [2006]

    Proceedings of the 23rd international conference on Machine learning (2006), 377–384

    Practical solutions to the problem of diagonal dominance in kernel document clustering. Proceedings of the 23rd international conference on Machine learning (2006), 377–384

    2006

  30. [2007]

    Proceedings of the 16th international conference on World Wide Web

    YAGO: a core of semantic knowledge. Proceedings of the 16th international conference on World Wide Web. (2007), 697–706. DOI:https://doi.org/10.1145/1242572.1242667

    2007

  31. [2008]

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2008), 207–218

    Real-time news event extraction for global crisis monitoring. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2008), 207–218

    2008

  32. [2009]

    Who, what, when, where, why?: comparing multiple approaches to the cross-lingual 5W task. Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP: Volume 1-Volume 1 (2009), 423–431

    2009

  33. [2010]

    (2009), 1–14

    English. (2009), 1–14

    2009

  34. [2011]

    Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing (2011), 782–792

    Robust Disambiguation of Named Entities in Text. Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing (2011), 782–792

    2011

  35. [2012]

    SUTime: A library for recognizing and normalizing time expressions. LREC. iii (2012), 3735–3740. DOI:https://doi.org/10.1017/CBO9781107415324.004

    2012 doi

  36. [2013]

    International Journal of Scientific & Engineering Research - IJSER

    News Event Extraction Using 5W1H Approach & Its Analysis. International Journal of Scientific & Engineering Research - IJSER. 4, 5 (2013), 2064–2067

    2013

  37. [2015]

    ICAICTA 2015 - 2015 International Conference on Advanced Informatics: Concepts, Theory and Applications (2015)

    Event extraction on Indonesian news article using multiclass categorization. ICAICTA 2015 - 2015 International Conference on Advanced Informatics: Concepts, Theory and Applications (2015)

    2015

  38. [2016]

    arXiv preprint arXiv:1605.07895

    Automatic Extraction of Causal Relations from Natural Language Texts: A Comprehensive Survey. arXiv preprint arXiv:1605.07895. (2016)

    2016 arXiv

  39. [2017]

    Proceedings of the 15th International Symposium of Information Science (2017), 218–223

    news-please: A Generic News Crawler and Extractor. Proceedings of the 15th International Symposium of Information Science (2017), 218–223

    2017

  40. [2018]

    International Journal on Digital Libraries

    Automated identification of media bias in news articles: an interdisciplinary literature review. International Journal on Digital Libraries. (2018), 1–25. DOI:https://doi.org/10.1007/s00799-018-0261-y

    2018 doi

  41. [2019]

    In Proceedings of the 7th International Workshop on News Recommendation and Analytics, Copen-hagen, Denmark, September 2019 (INRA 2019), 8 pages

    Giveme5W1H: A Universal Sys-tem for Extracting Main Events from News Articles. In Proceedings of the 7th International Workshop on News Recommendation and Analytics, Copen-hagen, Denmark, September 2019 (INRA 2019), 8 pages. 1 INTRODUCTION The extraction of a news article’s ma...

    2019

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.