Evaluating the Robustness of Embedding-based Topic Models to OCR Noise

Elaine Zosa; Mark Granroth-Wilding; Stephen Mutuvi; Antoine Doucet

doi:10.1007/978-3-030-91669-5_30

Evaluating the Robustness of Embedding-based Topic Models to OCR Noise

Elaine Zosa, Mark Granroth-Wilding, Stephen Mutuvi, Antoine Doucet

Tutkimustuotos: Artikkeli kirjassa/raportissa/konferenssijulkaisussa › Konferenssiartikkeli › Tieteellinen › vertaisarvioitu

Abstrakti

Unsupervised topic models such as Latent Dirichlet Allocation (LDA) are popular tools to analyse digitised corpora. However, the performance of these tools have been shown to degrade with OCR noise. Topic models that incorporate word embeddings during inference have been proposed to address the limitations of LDA, but these models have not seen much use in historical text analysis. In this paper we explore the impact of OCR noise on two embedding-based models, Gaussian LDA and the Embedded Topic Model (ETM) and compare their performance to LDA. Our results show that these models, especially ETM, are slightly more resilient than LDA in the presence of noise in terms of topic quality and classification accuracy.

Alkuperäiskieli	englanti
Otsikko	Towards Open and Trustworthy Digital Societies. ICADL 2021
Toimittajat	Hao-Ren Ke, Chei Sian Lee, Kazunari Sugiyama
Sivumäärä	9
Julkaisupaikka	Cham
Kustantaja	Springer
Julkaisupäivä	30 marrask. 2021
Sivut	392-400
ISBN (painettu)	978-3-030-91668-8
ISBN (elektroninen)	978-3-030-91669-5
DOI - pysyväislinkit	https://doi.org/10.1007/978-3-030-91669-5_30
Tila	Julkaistu - 30 marrask. 2021
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisuussa
Tapahtuma	International Conference on Asia-Pacific Digital Libraries - online Kesto: 1 jouluk. 2021 → 3 jouluk. 2021 Konferenssinumero: 23 https://icadl.net/icadl2021/

Julkaisusarja

Nimi	Lecture Notes in Computer Science
Vuosikerta	13133
ISSN (painettu)	0302-9743
ISSN (elektroninen)	1611-3349

Tieteenalat

113 Tietojenkäsittely- ja informaatiotieteet

Pääsy asiakirjaan

10.1007/978-3-030-91669-5_30

ICADL_OCR_impactHyväksytty kirjoittajan käsikirjoitus, 361 KBLisenssi: Määrittelemätön

Siteeraa tätä

Zosa, E., Granroth-Wilding, M., Mutuvi, S., & Doucet, A. (2021). Evaluating the Robustness of Embedding-based Topic Models to OCR Noise. teoksessa H-R. Ke, C. S. Lee, & K. Sugiyama (Toimittajat), Towards Open and Trustworthy Digital Societies. ICADL 2021 (Sivut 392-400). (Lecture Notes in Computer Science; Vuosikerta 13133). Springer. https://doi.org/10.1007/978-3-030-91669-5_30

@inproceedings{429ea864336b481387203726c13c828f,

title = "Evaluating the Robustness of Embedding-based Topic Models to OCR Noise",

abstract = "Unsupervised topic models such as Latent Dirichlet Allocation (LDA) are popular tools to analyse digitised corpora. However, the performance of these tools have been shown to degrade with OCR noise. Topic models that incorporate word embeddings during inference have been proposed to address the limitations of LDA, but these models have not seen much use in historical text analysis. In this paper we explore the impact of OCR noise on two embedding-based models, Gaussian LDA and the Embedded Topic Model (ETM) and compare their performance to LDA. Our results show that these models, especially ETM, are slightly more resilient than LDA in the presence of noise in terms of topic quality and classification accuracy.",

keywords = "113 Computer and information sciences, topic modelling, word embeddings, OCR",

author = "Elaine Zosa and Mark Granroth-Wilding and Stephen Mutuvi and Antoine Doucet",

year = "2021",

month = nov,

day = "30",

doi = "10.1007/978-3-030-91669-5_30",

language = "English",

isbn = "978-3-030-91668-8",

series = "Lecture Notes in Computer Science",

publisher = "Springer",

pages = "392--400",

editor = "Ke, {Hao-Ren } and Lee, {Chei Sian} and Sugiyama, {Kazunari }",

booktitle = "Towards Open and Trustworthy Digital Societies. ICADL 2021",

address = "Switzerland",

note = "International Conference on Asia-Pacific Digital Libraries, ICADL ; Conference date: 01-12-2021 Through 03-12-2021",

url = "https://icadl.net/icadl2021/",

}

Zosa, E, Granroth-Wilding, M, Mutuvi, S & Doucet, A 2021, Evaluating the Robustness of Embedding-based Topic Models to OCR Noise. julkaisussa H-R Ke, CS Lee & K Sugiyama (toim), Towards Open and Trustworthy Digital Societies. ICADL 2021. Lecture Notes in Computer Science, Vuosikerta 13133, Springer, Cham, Sivut 392-400, International Conference on Asia-Pacific Digital Libraries, 01/12/2021. https://doi.org/10.1007/978-3-030-91669-5_30

Evaluating the Robustness of Embedding-based Topic Models to OCR Noise. / Zosa, Elaine; Granroth-Wilding, Mark; Mutuvi, Stephen et al.

Towards Open and Trustworthy Digital Societies. ICADL 2021. toim. / Hao-Ren Ke; Chei Sian Lee; Kazunari Sugiyama. Cham : Springer, 2021. s. 392-400 (Lecture Notes in Computer Science; Vuosikerta 13133).

Tutkimustuotos: Artikkeli kirjassa/raportissa/konferenssijulkaisussa › Konferenssiartikkeli › Tieteellinen › vertaisarvioitu

TY - GEN

T1 - Evaluating the Robustness of Embedding-based Topic Models to OCR Noise

AU - Zosa, Elaine

AU - Granroth-Wilding, Mark

AU - Mutuvi, Stephen

AU - Doucet, Antoine

N1 - Conference code: 23

PY - 2021/11/30

Y1 - 2021/11/30

N2 - Unsupervised topic models such as Latent Dirichlet Allocation (LDA) are popular tools to analyse digitised corpora. However, the performance of these tools have been shown to degrade with OCR noise. Topic models that incorporate word embeddings during inference have been proposed to address the limitations of LDA, but these models have not seen much use in historical text analysis. In this paper we explore the impact of OCR noise on two embedding-based models, Gaussian LDA and the Embedded Topic Model (ETM) and compare their performance to LDA. Our results show that these models, especially ETM, are slightly more resilient than LDA in the presence of noise in terms of topic quality and classification accuracy.

AB - Unsupervised topic models such as Latent Dirichlet Allocation (LDA) are popular tools to analyse digitised corpora. However, the performance of these tools have been shown to degrade with OCR noise. Topic models that incorporate word embeddings during inference have been proposed to address the limitations of LDA, but these models have not seen much use in historical text analysis. In this paper we explore the impact of OCR noise on two embedding-based models, Gaussian LDA and the Embedded Topic Model (ETM) and compare their performance to LDA. Our results show that these models, especially ETM, are slightly more resilient than LDA in the presence of noise in terms of topic quality and classification accuracy.

KW - 113 Computer and information sciences

KW - topic modelling

KW - word embeddings

KW - OCR

U2 - 10.1007/978-3-030-91669-5_30

DO - 10.1007/978-3-030-91669-5_30

M3 - Conference contribution

SN - 978-3-030-91668-8

T3 - Lecture Notes in Computer Science

SP - 392

EP - 400

BT - Towards Open and Trustworthy Digital Societies. ICADL 2021

A2 - Ke, Hao-Ren

A2 - Lee, Chei Sian

A2 - Sugiyama, Kazunari

PB - Springer

CY - Cham

T2 - International Conference on Asia-Pacific Digital Libraries

Y2 - 1 December 2021 through 3 December 2021

ER -