A Comprehensive Study on the Use of Word Embedding Models in Software Engineering Domain

Jingxuan Zhang; Lianyi Zhi; Qianshuang Meng; Weiqin Zou; Xiaohan Chen

REVIEW 5 major objections 6 minor 32 references

A Comprehensive Study on the Use of Word Embedding Models in Software Engineering Domain

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This systematic review of 181 studies finds word embeddings in software engineering concentrated in maintenance and development tasks, led by Word2Vec and BERT, and rarely compared against traditional baselines.

desk verdict Useful first map of word-embedding use in SE, but the corpus rests on an impossible DBLP full-text search and the model-popularity ranking is partly a search artifact. read the letter →

arxiv 2505.17634 v1 pith:GV4VUOGH submitted 2025-05-23 cs.SE

Xiaohan Chen , Weiqin Zou , Lianyi Zhi , Qianshuang Meng , Jingxuan Zhang This is my paper · ORCID

classification cs.SE

keywords wordembeddingsoftwareengineeringsystematicliteraturereviewWord2VecBERTsemanticrepresentationartifactstaxonomy

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

This paper is a systematic literature review asking how word embedding (WE) models are actually used inside software engineering research. The authors searched 45 CCF A/B-ranked SE venues with six keywords, filtered to 181 primary studies, and coded them by task, artifact, model, comparison practice, and training data. They claim that adoption rose year by year through 2022 with a dip in 2023; that maintenance and development tasks dominate; that Word2Vec and BERT are the most used models; that SE-specific embeddings are preferred over generic pre-trained ones; and that the field lacks systematic comparison of embeddings against each other or against simpler baselines. If these descriptive claims hold, they give the SE community a map of where embeddings already help and where the evidence for their value is thinnest.

What carries the argument

The machinery is the review protocol itself. The authors searched DBLP for six exact keywords ('word embedding', 'word2vec', 'GloVe', 'fasttext', 'BERT', 'ELMo') across 45 CCF A/B SE venues, applied inclusion/exclusion criteria to get 181 primary studies, then used open coding and closed and open card sorting to build a taxonomy of SE tasks organized by software lifecycle areas (requirements, development, testing, maintenance) plus two emergent categories (general task support, project management). Four research questions (RQ1-RQ4) section the analysis by publication trend, task/artifact distribution, model choice and comparison practice, and training-data strategy.

What would settle it

Re-run the same search over the same 45 venues and years using full-text databases (e.g., ACM Digital Library, IEEE Xplore), keeping the six keywords and adding unlisted model names such as 'CodeBERT', 'RoBERTa', and 'embedding layer' as well as synonym phrases like 'vector representation'; if the primary-study count rises materially or the relative shares of Word2Vec, BERT, and SE-specific training shift, the paper's headline distributions are incomplete.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is a descriptive landscape: of 181 primary studies, software maintenance (36%) and development (24%) account for most uses; defect handling and code recommendation/generation are the dominant subareas; Word2Vec appears in 98 studies and BERT/variants in 54, far ahead of GloVe, FastText, and ELMo; 118 studies trained or fine-tuned SE-specific embeddings while 40 used generic ones; and 84% of studies never compared their chosen embedding against traditional representations such as TF-IDF, while only 22% ran direct experimental comparisons between different WE models. The authors present this as evidence that the field has adopted embeddings enthusiastically but without the comparative evaluation needed to justify model choice.

Load-bearing premise

The whole survey rests on the assumption that the six-keyword DBLP search finds every relevant paper; since DBLP indexes bibliographic metadata rather than full text, studies that use a word embedding but do not literally contain one of those six phrases can fall out of the 181-study set and change the reported distributions.

Editorial extensions

If this is right

Researchers entering SE tasks can use the taxonomy to see which areas already have embedding practice and which (requirements engineering, testing) are comparatively open.
The 84% no-comparison rate means most reported gains of embeddings over simple baselines are not empirically demonstrated, so the field needs standard baselines such as TF-IDF in future evaluations.
With 118 of 181 studies using SE-specific data, the de facto standard is to train or fine-tune on SE corpora, not to plug in generic pre-trained vectors.
The 2023 decline and the rise of seq2seq/LLM approaches imply that static word embeddings are being folded into larger context-aware representation models.
The concentration in TSE, ICSE, and ASE tells authors where this cross-disciplinary work is currently published and accepted.

Reading between the lines

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

The six-keyword search likely undercounts studies that name models outside the list, so the Word2Vec/BERT ranking is best read as a lower bound; a full-text search over the same venues could shift the proportions.
Only 15 studies directly compared SE-specific and general embeddings, so the paper's 'generally favored' conclusion describes prevailing practice rather than demonstrated superiority.
Applying the same taxonomy to 2024-2025 publications would test whether the maintenance/development dominance survives the rise of LLM-based context-aware representations.
The 2023 dip is plausibly the leading edge of a shift from standalone static embeddings to embedding layers inside seq2seq and LLM pipelines, a shift that would make the survey's model counts an endpoint of an era.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Read the paper. The useful core is the taxonomy of SE tasks that use word embeddings and the descriptive statistics about models and training strategies. If you are new to the area, this gives a reasonable first map of where embeddings have been applied and what practices are common. The coding process (two authors, open card sorting, discussion) is a legitimate way to build a taxonomy, and the gap analysis — most studies never compare embeddings against TF-IDF or against each other — is a real and useful observation.

The soft spots are serious. The search strategy in Section 2.1.2 says they searched DBLP for papers whose full-text contains one of six keywords. DBLP indexes bibliographic metadata, not full text, so that search cannot work as described. That means the 1,957 candidates and the 181 primary studies were almost certainly retrieved by metadata matches, which systematically misses papers that use embeddings without the exact strings in the title or indexed fields. The model-popularity result in RQ3 is particularly exposed: Word2Vec and BERT are among the six search keywords, so their leading position is partly a selection artifact. The paper also reports 181 primary studies in the abstract and results but 156 in the conclusion, and the full list of primary studies is not in the paper — it is on an external spreadsheet. I cannot verify the counts or the taxonomy without that list, and the discrepancy suggests the dataset was not finalized.

The circularity point is not a refutation of every finding. The task taxonomy and the artifact distribution probably survive a recall bias in the search; the claim that maintenance and development dominate is plausible from broader knowledge of the field. But the paper's quantitative claims — trends, model rankings, comparison rates — should not be taken at face value.

Who is this for? Researchers who want a starting bibliography and a task-level map of embedding use in SE. That value is real but modest, and it is undermined by the unreliability of the corpus construction. I would send it to review, not desk-reject: a serious referee can push the authors to fix the search description, reconcile the counts, publish the primary study list, and either drop or re-frame the model-ranking claims. After that, it could be a citable reference. As is, I would not cite the numbers.

Referee Report

5 major / 6 minor

Summary. The paper reports a systematic literature review of word-embedding (WE) model use in software engineering. The authors describe searching 45 CCF A/B-ranked SE venues using six keywords, screening 1,957 candidate papers, and retaining 181 primary studies. Four research questions address publication trends, SE tasks and artifacts, the popularity of specific WE models, and whether models are trained from scratch, fine-tuned, or used as generic pre-trained embeddings. The main findings are that WE adoption rose until 2022 with a decline in 2023, software maintenance and development dominate, Word2Vec and BERT are the most frequently used models, SE-specific embeddings are generally preferred, and comparative evaluations are rare. The manuscript also proposes a taxonomy of SE tasks organized around the software lifecycle.

Significance. If the review were methodologically sound, it would be a useful reference for SE researchers selecting embedding models and for identifying gaps in comparative evaluation. The manual classification of 181 papers into a task taxonomy and the artifact-level analysis are potentially valuable contributions. However, as presented, the search strategy is not reproducible as described, the model-popularity ranking is partly a consequence of the chosen search keywords, and several reported numbers are internally inconsistent. These issues currently undermine the reliability of the main claims; with a corrected and transparent search and reconciled data, the paper could make a meaningful contribution.

major comments (5)

[§2.1.2] The search strategy states that DBLP was used to retrieve papers 'whose full-text contains at least one keyword.' DBLP is a bibliographic metadata database and does not provide full-text search over paper contents, so the described procedure cannot retrieve all papers containing the keywords in their full text. This is load-bearing because the 1,957 candidate set and the 181 primary studies are the sole evidence for all four research questions. The authors should either describe a reproducible full-text search over an appropriate database (e.g., ACM Digital Library, IEEE Xplore, Scopus, or a provided search API with exact query strings and dates) or rerun the search and check whether the findings change.
[§2.1.2 and §3.3.1] The set of search keywords includes the model names 'word2vec', 'GloVe', 'fasttext', 'BERT', and 'ELMo'. Since RQ3 concludes that Word2Vec and BERT are the most common models, the ranking is in part a selection effect: papers whose metadata or full text contains these exact strings are more likely to enter the candidate set, while papers that mention only 'CodeBERT', 'embedding layer', or 'neural embedding' may be missed unless they also contain one of the chosen keywords. The authors should broaden the keyword set or perform a sensitivity analysis with additional terms (for example, 'embedding', 'CodeBERT', 'pre-trained embedding', 'neural embedding') and report the resulting change in the corpus and model distribution.
[Abstract, §1, and §7] The number of primary studies is inconsistent: the abstract and introduction state 181, while the conclusion states 156. This discrepancy is not explained and makes it impossible to know the actual size of the dataset. The authors must reconcile these numbers and ensure that all subsequent counts, including the percentages and the taxonomy tables, agree with a single final corpus.
[§3.4, Figure 4] The RQ4 statistics do not add up. The text reports 9 papers with unclear generation details, 118 papers using SE-specific embeddings, 40 papers using generic embeddings, and 16 papers using both, which sums to 183, not 181. In addition, within the SE-specific group, 83 papers trained from scratch and 33 fine-tuned sum to 116, not 118. These arithmetic inconsistencies must be corrected, and the categories should be mutually exclusive and exhaustive with respect to the 181 primary studies.
[§3.3.1, Figure 3] The model usage counts in Figure 3 (Word2Vec 98, BERT 54, GloVe 22, FastText 21, ELMo 2) sum to 197, which exceeds 181. If a primary study can use more than one WE model, the paper should state this explicitly and present per-model versus per-study counts separately. As written, the reader cannot tell whether the frequencies are numbers of papers or numbers of usages, which directly affects the interpretation of RQ3.

minor comments (6)

[§2.1.1] The list of 45 selected venues is not included in the manuscript; the authors point to an external spreadsheet link. For a systematic review, the venue list and the complete list of primary studies should be provided in an appendix, supplementary material, or a permanent repository, so that the reader can verify the corpus and reproduce the taxonomy.
[§3.1] The text refers to Figure 1 for the year and venue distribution but does not report the underlying counts in the text or a table. Adding a table with the number of primary studies per year and per venue would improve verifiability.
[§3.4] The sentence 'it would be interesting and very valuable to conduct further investigation' follows the RQ4 summary; the wording should be adjusted to avoid an ungrammatical shift from the preceding discussion.
[§5.1] The internal validity discussion acknowledges that not all papers were read in full, but it does not address the more direct threats to the search: the impossibility of the described DBLP full-text search and the bias introduced by using model names as search keywords. These threats should be acknowledged and mitigated.
[References] Reference [14], cited as the first study applying Word2Vec in SE, is titled 'Mining user opinions in mobile app reviews: A keyword-based approach'; the connection to Word2Vec should be clarified or the citation corrected.
[Throughout] Several typographical and formatting issues remain, including inconsistent spacing and 'V olume' in the reference list; these should be cleaned up in a final pass.

Circularity Check

1 steps flagged · score 5.0 of 10

RQ3's model-popularity ranking is partly an artifact of the DBLP keyword search that included Word2Vec and BERT; the taxonomy and trend analyses are not circular.

self definitional [Section 2.1.2 (Keywords) and Section 3.3.1 (Distribution of various WE models used in SE tasks)]
"In addition to the keyword “word embedding”, we also add the names of existing WE models as search keywords, namely “word2vec”, “GloVe”, “fasttext”, “BERT” and “ELMo”. ... Word2Vec (98 related studies) is clearly the most commonly used word embedding model ... Following Word2Vec, BERT (including its variants, such as CodeBERT) ranked second in usage, with 54 studies utilizing this bidirectional transformer model."

The candidate pool is created by querying DBLP for exactly the model names that later dominate the RQ3 frequency table. Because “word2vec” and “BERT” are search keywords, and “BERT” also matches “CodeBERT” as a substring, papers whose indexed metadata name Word2Vec or BERT are systematically admitted, while models not in the keyword list (e.g., Doc2Vec, Sent2Vec, task-specific embeddings) can enter only if they also happen to mention “word embedding” or one of the six terms. The reported ranking—Word2Vec 98 and BERT 54—is therefore not an independent measurement of model popularity in the SE literature; it is partly generated by the same terms used to define the corpus.

full rationale

The review's main non-circular content is the manual taxonomy (RQ2) and the descriptive year/venue trend (RQ1), which are coded from the collected papers and do not reduce to the search terms. RQ4's generic-vs-SE-specific distinction is also independently coded. The one genuine circular burden sits on RQ3: Section 2.1.2 constructs the candidate pool with six keywords, five of which are exact model names ('word2vec', 'GloVe', 'fasttext', 'BERT', 'ELMo'), and Section 3.3.1 then reports the frequency of those same model names as an empirical finding. Because DBLP does not provide full-text search and the paper claims it does, the retrieval is limited to bibliographic metadata, so papers that use a WE model without one of these six strings in metadata are missed. This makes the Word2Vec/BERT ranking partly a selection effect, not an independent empirical result. The 181-vs-156 primary-study inconsistency (abstract vs conclusion) reinforces that the corpus is not fully reproducible; that is a correctness issue rather than a circular one. No equations or fitted parameters are involved elsewhere, and no load-bearing self-citation appears.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The review introduces no new entities or parameters; its conclusions rest on the search and screening methodology, which are domain assumptions about coverage and coding reliability.

assumptions (3)

domain assumption The six search keywords ('word embedding', 'word2vec', 'GloVe', 'fasttext', 'BERT', 'ELMo') and the 45 CCF A/B SE venues capture all relevant WE-in-SE studies.
The review's counts and distributions depend on this. DBLP does not index full text, so papers using WE without these keywords may be missed. Section 2.1.2.
domain assumption Manual title/abstract screening by two authors reliably identifies relevant studies and correctly classifies SE tasks and artifacts.
The taxonomy and counts depend on subjective coding. Section 2.1.3 and 3.2.1.
domain assumption CCF A/B ranked venues are a representative sample of the SE research field.
The review excludes C-ranked venues and gray literature, which could bias the trends. Section 5.2.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Comprehensive Study on the Use of Word Embedding Models in Software Engineering Domain." pith.science (2026). https://pith.science/paper/GV4VUOGH

@misc{pith2026250517634,
  author       = {Pith},
  title        = {Pith review of: A Comprehensive Study on the Use of Word Embedding Models in Software Engineering Domain},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GV4VUOGH}},
  note         = {Machine review of arXiv:2505.17634}
}

read the original abstract

Word embedding (WE) techniques are advanced textual semantic representation models oriented from the natural language processing (NLP) area. Inspired by their effectiveness in facilitating various NLP tasks, more and more researchers attempt to adopt these WE models for their software engineering (SE) tasks, of which semantic representation of software artifacts such as bug reports and code snippets is the basis for further model building. However, existing studies are generally isolated from each other without comprehensive comparison and discussion. This not only makes the best practice of such cross-discipline technique adoption buried in scattered papers, but also makes us kind of blind to current progress in the semantic representation of SE artifacts. To this end, we decided to perform a comprehensive study on the use of WE models in the SE domain. 181 primary studies published in mainstream software engineering venues are collected for analysis. Several research questions related to the SE applications, the training strategy of WE models, the comparison with traditional semantic representation methods, etc., are answered. With the answers, we get a systematical view of the current practice of using WE for the SE domain, and figure out the challenges and actions in adopting or developing practical semantic representation approaches for the SE artifacts used in a series of SE tasks.

Figures

Figures reproduced from arXiv: 2505.17634 by the authors.

**Figure 2.** The distribution of software artifacts that use WE models for semantic representation in all 181 papers. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

**Figure 3.** The distribution of different WE models in SE. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: The type and training methods of WE in SE. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 31 canonical work pages

[1]

Impact of word embedding models on text analytics in deep learning environment: a review

Asudani S D, Nagwani N K, Singh P. Impact of word embedding models on text analytics in deep learning environment: a review. Artif Intell Rev, 2023, 56: 10345–10425

work page 2023
[2]

An empirical study on the potential of word embedding techniques in bug report management tasks

Chen B T, Zou W Q, Cai B Y , et al. An empirical study on the potential of word embedding techniques in bug report management tasks. Empir Softw Eng, 2024, 29: 122

work page 2024
[3]

Linguistic regularities in continuous space word representations

Mikolov T, Yih W-t, Zweig G. Linguistic regularities in continuous space word representations. In: Vanderwende L., Daumé H. III, Kirchhoff K., eds. Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Atlanta, Georgia, 2013. 746–751

work page 2013
[4]

A review on word embedding techniques for text classification

Selva Birunda S, Kanniga Devi R. A review on word embedding techniques for text classification. In: Raj J. S., Iliyasu A. M., Bestak R., Baig Z. A., eds. Innovative Data Communication Technologies and Application, Singapore, 2021. 267–281

work page 2021
[5]

Deep learning with word embeddings improves biomedical named entity recognition

Habibi M, Weber L, Neves M, et al. Deep learning with word embeddings improves biomedical named entity recognition. Bioinformatics, 2017. i37–i48

work page 2017
[6]

Learning continuous word embedding with metadata for question retrieval in community question answering

Zhou G, He T, Zhao J, et al. Learning continuous word embedding with metadata for question retrieval in community question answering. In: Zong C, Strube M, eds. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (V olume 1: Long Papers), Beijing,...

work page 2015
[7]

Deep contextualized Word representations

Peters M E, Neumann M, Iyyer M, et al. Deep contextualized Word representations. In: Walker M, Ji H, Stent A, eds. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long Papers), New Orleans, Louisiana, 2018. 2227–2237

work page 2018
[8]

Bert: Pre-training of deep bidirectional transformers for language understand- ing

Devlin J, Chang M-W., Lee K, et al. Bert: Pre-training of deep bidirectional transformers for language understand- ing. In: Burstein J, Doran C, Solorio T, eds. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long and Short Papers), Minneapolis, Minn...

work page 2019

Show all 32 references

[9]

Software development methods: Review and outlook

Ma X-X, Liu X-Z, Xie B, et al. Software development methods: Review and outlook. J Softw, 2019, 30: 3–21 17

2019
[10]

FineLocator: A novel approach to method-level fine-grained bug localization by query expansion

Zhang W, Li Z, Wang Q, et al. FineLocator: A novel approach to method-level fine-grained bug localization by query expansion. Inform Software Tech, 2019, 110: 121–135

2019
[11]

Fast changeset-based bug localization with BERT

Ciborowska A, Damevski K. Fast changeset-based bug localization with BERT. Proceedings of the 44th Interna- tional Conference on Software Engineering, New York, NY , USA, 2022. 946–957

2022
[12]

Automatic text input generation for mobile testing

Liu P, Zhang X, Pistoia M, et al. Automatic text input generation for mobile testing. Proceedings of the 39th International Conference on Software Engineering, Buenos Aires, Argentina, 2017. 643-653

2017
[13]

Apiro: A framework for automated security tools api recommendation

Sworna Z T., Islam C, Babar M A. Apiro: A framework for automated security tools api recommendation. ACM T Softw Eng Meth, 2023, 32: 1–42

2023
[14]

Mining user opinions in mobile app reviews: A keyword-based approach (t)

Phong M V , Nguyen T T, Pham H V , et al. Mining user opinions in mobile app reviews: A keyword-based approach (t). Proceedings of the 30th International Conference on Automated Software Engineering, Lincoln, NE, USA,2015. 749–759

2015
[15]

Efficient estimation of word representations in vector space

Mikolov T, Chen K, Corrado G, et al. Efficient estimation of word representations in vector space. In: Bengio Y , LeCun Y , eds. Workshop Track Proceedings of the 1st International Conference on Learning Representations, Scottsdale, Arizona, USA, 2013

2013
[16]

Glove: Global vectors for word representation

Pennington J, Socher R, Manning C D. Glove: Global vectors for word representation. In: Moschitti A, Pang B, Daelemans W, eds. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, Doha, Qatar, 2014. 1532–1543

2014
[17]

Bag of tricks for efficient text classification

Joulin A, Grave E, Bojanowski P, et al. Bag of tricks for efficient text classification. In: Lapata M, Blunsom P, Koller A, eds. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, Valencia, Spain, 2017. 427–431

2017
[18]

Grounded theory research: Procedures, canons, and evaluative criteria

Corbin J M, Strauss A. Grounded theory research: Procedures, canons, and evaluative criteria. Qual Sociol, 1990, 13: 3–21

1990
[19]

Machine learning for software engineering: A tertiary study

Kotti Z, Galanopoulou R, Spinellis D. Machine learning for software engineering: A tertiary study. ACM Comput Surv, 2023, 55: 1–39

2023
[20]

A survey on deep learning for software engineering

Yang Y , Xia X, Lo D, et al. A survey on deep learning for software engineering. ACM Comput Surv, 2022, 54: 1–73

2022
[21]

Trends in software engineering processes using deep learning: a systematic literature review

Del Carpio A F, Angarita L B. Trends in software engineering processes using deep learning: a systematic literature review. Proceedings of the 46th Euromicro Conference on Software Engineering and Advanced Applications, Portoroz, Slovenia, 2020. 445–454

2020
[22]

Machine/deep learning for software engineering: A systematic literature review

Wang S, Huang L, Gao A, et al. Machine/deep learning for software engineering: A systematic literature review. IEEE T Software Eng, 2022, 49: 1188–1231

2022
[23]

Large language models for software engineering: A systematic literature review

Hou X, Zhao Y , Liu Y , et al. Large language models for software engineering: A systematic literature review. CoRR, 2023, abs/2308.10620

2023 arXiv
[24]

Text classification using embeddings: a survey

da Costa L S, Oliveira I L, Fileto R. Text classification using embeddings: a survey. Knowl Inf Syst, 2023, 65: 2761–2803

2023
[25]

A review on word embedding techniques for text classification

Selva Birunda S, Kanniga Devi R. A review on word embedding techniques for text classification. In: Raj J S, Iliyasu A M, Bestak R, Baig Z A, eds. Innovative Data Communication Technologies and Application, Singapore,
[26]

A systematic literature review on word embeddings

Gutiérrez L, Keith B. A systematic literature review on word embeddings. In: Mejia J, Muñoz M, Rocha Á, Peña A, Pérez-Cisneros M, eds. Trends and Applications in Software Engineering, Cham, 2019. 132–141

2019
[27]

A detailed review on word embedding techniques with emphasis on word2vec

Johnson S J, Murty M R, Navakanth I. A detailed review on word embedding techniques with emphasis on word2vec. Multimed Tools Appl, 2024, 83: 37979–38007

2024
[28]

From word to sense embeddings: A survey on vector representations of meaning

Camacho-Collados J, Pilehvar M T. From word to sense embeddings: A survey on vector representations of meaning. J Artif Intell Res, 2018, 63: 743–788

2018
[29]

A comprehensive survey on aspect based word embedding models and sentiment analysis classification approaches

Agrawal M, Moparthi N R. A comprehensive survey on aspect based word embedding models and sentiment analysis classification approaches. Recent Trends in Intensive Computing, 2021, 39: 33–39

2021
[30]

Word embeddings for sentiment analysis: a comprehensive empirical survey

Çano E, Morisio M. Word embeddings for sentiment analysis: a comprehensive empirical survey. CoRR, 2019, abs/1902.00753

2019 arXiv
[31]

A survey of cross-lingual word embedding models

Ruder S, Vuli´c I, Søgaard A. A survey of cross-lingual word embedding models. J Artif Intell Res, 2019, 65: 569–631

2019
[32]

Beyond word embeddings: A survey

Incitti F, Urli F, Snidaro L. Beyond word embeddings: A survey. Inform Fusion, 2023, 89: 418–436 18

2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Impact of word embedding models on text analytics in deep learning environment: a review

Asudani S D, Nagwani N K, Singh P. Impact of word embedding models on text analytics in deep learning environment: a review. Artif Intell Rev, 2023, 56: 10345–10425

work page 2023

[2] [2]

An empirical study on the potential of word embedding techniques in bug report management tasks

Chen B T, Zou W Q, Cai B Y , et al. An empirical study on the potential of word embedding techniques in bug report management tasks. Empir Softw Eng, 2024, 29: 122

work page 2024

[3] [3]

Linguistic regularities in continuous space word representations

Mikolov T, Yih W-t, Zweig G. Linguistic regularities in continuous space word representations. In: Vanderwende L., Daumé H. III, Kirchhoff K., eds. Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Atlanta, Georgia, 2013. 746–751

work page 2013

[4] [4]

A review on word embedding techniques for text classification

Selva Birunda S, Kanniga Devi R. A review on word embedding techniques for text classification. In: Raj J. S., Iliyasu A. M., Bestak R., Baig Z. A., eds. Innovative Data Communication Technologies and Application, Singapore, 2021. 267–281

work page 2021

[5] [5]

Deep learning with word embeddings improves biomedical named entity recognition

Habibi M, Weber L, Neves M, et al. Deep learning with word embeddings improves biomedical named entity recognition. Bioinformatics, 2017. i37–i48

work page 2017

[6] [6]

Learning continuous word embedding with metadata for question retrieval in community question answering

Zhou G, He T, Zhao J, et al. Learning continuous word embedding with metadata for question retrieval in community question answering. In: Zong C, Strube M, eds. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (V olume 1: Long Papers), Beijing,...

work page 2015

[7] [7]

Deep contextualized Word representations

Peters M E, Neumann M, Iyyer M, et al. Deep contextualized Word representations. In: Walker M, Ji H, Stent A, eds. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long Papers), New Orleans, Louisiana, 2018. 2227–2237

work page 2018

[8] [8]

Bert: Pre-training of deep bidirectional transformers for language understand- ing

Devlin J, Chang M-W., Lee K, et al. Bert: Pre-training of deep bidirectional transformers for language understand- ing. In: Burstein J, Doran C, Solorio T, eds. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long and Short Papers), Minneapolis, Minn...

work page 2019