Similarity Learning for Authorship Verification in Social Media

Benedikt Boenninghoff; Dorothea Kolossa; Robert M. Nickel; Steffen Zeiler

REVIEW 3 major objections 5 minor 47 references

Similarity Learning for Authorship Verification in Social Media

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 A hierarchical recurrent Siamese network with a two-margin contrastive loss verifies authorship of short social-media texts at 83–85% accuracy, about 12–14 points above a character-n-gram baseline.

desk verdict Solid engineering result with an unresolved author-leakage question; the 12-point improvement over one baseline needs an author-disjoint split to be trusted. read the letter →

arxiv 1908.07844 v1 pith:EQIR6KZU submitted 2019-08-20 cs.CL cs.LGstat.ML

Benedikt Boenninghoff , Robert M. Nickel , Steffen Zeiler , Dorothea Kolossa This is my paper · ORCID

classification cs.CLcs.LGstat.ML

keywords authorshipverificationsimilaritylearningSiamesenetworkhierarchicalrecurrentneuralcontrastivelossdocumentembeddingssocialmediaforensicsLSTM

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 claims that a two-level recurrent Siamese network, trained with a margin-based contrastive loss, can tell whether two short social-media texts share an author far better than traditional character-n-gram methods. On a public benchmark assembled from novels, tweets, reviews, blog posts, emails and forum posts, the proposed system reaches 83.2% verification accuracy and 85.3% with simple data augmentation, against 70.9% for the 2016 baseline it re-implements. The claim matters because forensic authorship verification on social media is normally hamstrung by short texts, mixed genres and shifting topics, exactly where n-gram features fail. If the result holds, learned document embeddings, rather than hand-crafted style features, are a viable route to such verification, and the gap is large enough to matter in practice.

What carries the argument

The load-bearing object is the two-level hierarchical LSTM inside a Siamese architecture. The lower LSTM turns the word sequence of each sentence into a fixed sentence vector; the upper LSTM turns the sentence vectors into a fixed document embedding. Two such encoders share weights, and a Euclidean distance between the two document embeddings is judged against the threshold $(\tau_1+\tau_2)/2$. The modified contrastive loss, with separate margins for same-author and different-author pairs, is what makes the embeddings resist topic and genre shifts; fixed padding to maximum word and sentence counts makes the network trainable by backpropagation through time.

What would settle it

Inspect the fold construction used for the 10-fold cross-validation: if any author's texts occur in both training and test folds, rerun the comparison on author-disjoint folds and see whether the 83.2–85.3% accuracies survive. A second check is to run the same network against two or more published verifiers on identical folds; if the 12-point gap shrinks or disappears, the claim of significant improvement is benchmark-dependent.

Watch

Extended reading notes

Core claim

The central claim is that a hierarchical recurrent Siamese network (HRSN), which encodes a document first sentence-by-sentence and then document-wide with two stacked LSTM encoders, learns document embeddings whose Euclidean distance separates same-author from different-author pairs. The separation is induced by a modified contrastive loss with two thresholds, one pulling same-author cross-topic pairs closer than a lower margin and one pushing different-author same-topic pairs beyond an upper margin. On the cross-genre corpus used here, the authors report a verification accuracy increase of about 12 percentage points over the implemented n-gram baseline, a further roughly 2 points from randomizing the concatenation order of known documents, and no gain from semi-supervised pretraining. The authors conclude that the topology and loss, rather than pretraining, carry the improvement.

Load-bearing premise

The load-bearing premise is that the benchmark corpus and its 10-fold splits estimate real forensic social-media performance without leakage, and that the single 2016 character-n-gram baseline is a representative yardstick; if texts by the same author appear in both training and test folds, or the baseline is unrepresentatively weak, the reported 12-point gain would be inflated.

Editorial extensions

If this is right

Same-author verification for short, mixed-genre social media texts can be automated at roughly 85% accuracy using learned embeddings alone.
The method needs no hand-crafted stylometric features, so it transfers to social-media genres where part-of-speech taggers and n-gram statistics are unreliable.
Data augmentation by randomizing the concatenation order of known documents adds about 2 points, so the model benefits from more pair variability without new data.
Semi-supervised pretraining on out-of-domain sentence-pair and paragraph data did not help, implying that in-domain training data is the bottleneck.
The gap over the n-gram baseline (70.9% versus 85.3%) suggests that earlier feature-based systems were not close to the practical ceiling on this task.

Reading between the lines

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

One testable extension is to report author-disjoint splits: the paper does not state whether the same author's texts can appear in both training and test folds, and if they can, part of the gain may be author identity memorization rather than style transfer.
The 85.3% figure is measured against a single 2016 baseline; a fairer statement of 'significantly improves' would need comparison with other neural or compression-based verifiers on the same folds.
Because cross-topic same-author pairs are explicitly trained to be close, the same architecture could be adapted to cross-genre attribution, not just verification, by embedding candidate authors' documents in the same space.
A direct probe of the learned embeddings, checking whether same-author topics cluster while different-author same-topic texts separate, would give a visual check that the loss does what the paper's illustration suggests.

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.

Desk Editor's Note

read the letter

Short version: this is a decent engineering paper on authorship verification for short social media texts. It reports a real-looking accuracy gain over a reproduced baseline, but the evaluation leaves a key question unanswered: whether the cross-validation splits separate authors. If they do not, the headline gain may be partly author memorization.

The actual contribution is modest but real: combining a two-level hierarchical LSTM encoder inside a Siamese network, with a two-threshold contrastive loss that explicitly handles cross-topic same-author and same-topic different-author pairs. That specific combination, applied to the Halvani/PAN social media corpus, appears to be new. The paper is also honest about what didn't work (pretraining) and describes the baseline reproduction well enough that the 70.9% matches the published 71.1%.

What the paper does well: it gives enough architectural and training detail to be credible, and the baseline comparison is fair to the baseline. The data augmentation trick—randomizing the order of concatenating known documents—is a nice, simple idea. The numbers are consistent across folds.

Soft spots, in order: first, the paper never says whether the 10-fold splits are author-disjoint. The data consists of instances with a known-author document set and a questioned document; if the same author shows up in both training and test folds, the network can learn author-specific style and the reported 12-point gain over Halvani could be inflated. I can't tell from the text whether they guarded against that, but the burden should be on them to demonstrate it. Second, they compare to a single 2016 baseline. That baseline is the dataset's origin, which is sensible, but one baseline makes 'significantly improves' a thinner claim. They also use the word 'significantly' without a significance test; given the standard deviations, a 12-point gap is probably significant in a practical sense, but the authors should say so properly. Third, no code is released, which limits independent verification.

If the author-disjoint issue is resolved and the split is clean, the contribution stands as a useful, reproducible method for a niche but practically relevant task. The paper deserves a serious referee—not for novelty, but for solid engineering and a clear result. I'd ask the authors to clarify the split, add at least one more recent baseline, and run a proper significance test before publication.

Referee Report

3 major / 5 minor

Summary. This paper addresses authorship verification for short social-media texts. The authors propose a hierarchical recurrent Siamese network (HRSN) in which word-level and sentence-level LSTMs encode each document into a fixed-length embedding, and a modified contrastive loss with two thresholds encourages same-author distances below τ1 and different-author distances above τ2. Experiments on the Halvani/PAN corpus use 10-fold cross-validation. The HRSN is reported at 83.2%±0.8 accuracy versus 70.9%±1.7 for the implemented 2016 Halvani baseline, rising to 85.3%±0.9 with concatenation-order augmentation. The paper claims a significant improvement and discusses possible pretraining. No code or data release is mentioned.

Significance. If the evaluation is valid, the result is a useful demonstration that representation learning can improve forensic authorship verification on short cross-genre social-media texts, with a clearly specified architecture. Strengths include the explicit network equations and training details, the reproduction of the Halvani baseline close to the original published level (70.9% versus approximately 71.1%), and falsifiable quantitative claims. The main limitations are in the evaluation protocol: author-disjointness of the folds is not stated, no significance test supports the word "significantly," and only a single 2016 baseline is compared. These gaps currently leave the strength of the central claim uncertain, but they are fixable within the scope of the manuscript.

major comments (3)

[Sections 3.1 and 3.3] The paper never states whether the 10-fold cross-validation splits are disjoint at the author level. Section 3.1 defines instances as (Dknown, dunknown, l) with multiple known documents per author, and Section 3.3 only specifies the 80/10/10 train/dev/test percentages. If the same author occurs in both training and test folds, the hierarchical LSTM can memorize author-specific style during training, and the Table 1 gain over the capacity-limited Halvani baseline could be inflated. Please state the split criterion explicitly; if the splits are not author-disjoint, re-run the evaluation with author-disjoint folds and report the resulting numbers, or explicitly restrict the claims to a closed-set setting.
[Section 3.4 and abstract] The word "significantly" is used in the abstract and in Section 3.4, but no statistical significance test is reported. The reported means and standard deviations over 10 folds do not by themselves establish that the difference is significant. Because the folds are paired, please report a paired test across folds (for example, a paired t-test or Wilcoxon signed-rank test) on per-fold accuracies, or a confidence interval on the accuracy difference, together with the test statistic and p-value.
[Section 3.2 and Table 1] The evaluation compares the proposed HRSN with only one baseline, the re-implemented Halvani et al. (2016) method. To support the general claim that the proposed topology "significantly improves" performance on challenging social-media data, at least one additional strong baseline is needed, such as a top PAN 2015 author-identification system or a more recent neural authorship-verification method. Without such a comparison, the reported 12-point improvement may be specific to the chosen baseline rather than indicative of a general advantage.

minor comments (5)

[Section 2.1, Eq. (10)] Equation (10) writes x(d) = h(w)_T(s), but according to Figure 2 and the surrounding text the document embedding should be the final sentence-level hidden state, x(d) = h(s)_T(s). Please correct this notation.
[Section 2.2, Eq. (13)] The decision threshold is written as τ = τ1 + τ2 / 2, which is ambiguous under standard operator precedence; it should be written as τ = (τ1 + τ2) / 2.
[Section 3.3] The text says T(w)=33 and T(s)=123 were chosen "to cover >90% of the tokens of a single document in average"; since T(w) is a word-per-sentence limit and T(s) is a sentence-per-document limit, please clarify the wording to state which quantity each threshold covers.
[Section 3.4 and conclusion] Section 3.4 states the accuracy increased "from around 71% to 83.2%", while the conclusion says "improvements of about 15 percentage points"; the 15-point figure is appropriate only for the augmented result (85.3% − 70.9% = 14.4 points), so please align the wording.
[Conclusion and references] There is a typo, "pretrainig", in the conclusion; also, reference [32] duplicates reference [25] (both are Sutskever et al., "Sequence to sequence learning with neural networks") and should be consolidated.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the HRSN's accuracy is measured on held-out folds against an independent baseline, not derived from its own training objective.

full rationale

This paper makes an empirical performance claim: a hierarchical recurrent Siamese network trained with a contrastive loss achieves 83.2% vs 70.9% baseline on the Halvani/PAN corpus. The derivation chain is training followed by held-out evaluation, with no fitted parameter renamed as a prediction. The contrastive loss in Eq. (12) explicitly encodes the same margin structure used by the decision rule in Eq. (13), but that is the standard design of metric-learning systems: the loss defines the training objective, and the reported accuracy measures generalization on a 10% held-out test slice per fold (Section 3.3). The baseline is an independent re-implementation of Halvani et al. [11] with performance comparable to the original paper's Table 7. The architecture borrows LSTM, hierarchical, and Siamese components from prior work (e.g., [16], [18], [23]), but none of those citations is by the present authors and none is invoked as a uniqueness theorem or as a substitute for the reported experiments. The 'document embeddings' terminology is a naming choice, not a renamed known result. The only substantive concern raised in review, namely whether the 10-fold splits are author-disjoint so that gains could reflect author memorization rather than generalization, is a validity and experimental-design question, not a circularity: it does not make the output equivalent to the input by construction. Therefore no circular step is present, and the score is 0.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a standard supervised metric-learning setup: labeled author pairs, a corpus-level benchmark, and tuned hyperparameters. No new physical or conceptual entities are introduced, and the only assumptions are domain-level: that the benchmark represents forensic social-media text, that the splits avoid author leakage, and that pretrained word embeddings preserve style signals.

free parameters (8)

Loss thresholds τ1 and τ2 = not reported
Predefined margins in the contrastive loss (Eq. 12) that control how close same-author pairs must be and how far different-author pairs must be; values are not given in the paper and were presumably tuned on the development set.
Decision threshold τ = τ = (τ1 + τ2)/2
The final accept/reject threshold in Eq. 13 is derived from the two loss margins, so it inherits any tuning of τ1 and τ2.
Sentence and document embedding dimensions Ds and Dd = Ds = 150, Dd = 75
Chosen by halving the GloVe dimension of 300; these sizes affect capacity and were selected without a stated search.
Maximum word count T(w) and sentence count T(s) = T(w) = 33, T(s) = 123
Chosen to cover more than 90% of tokens and sentences per document; affects how variable-length texts are truncated.
Dropout rate = 0.3
Variational dropout rate for both LSTM levels, a tuned hyperparameter.
Initial learning rate for Adadelta = 1.0
Selected after noticing it yields the best results, as stated in Section 3.3.
Gradient clipping threshold = 5
Gradient norm cap used during training.
LSTM parameter initialization range = uniform(-0.05, 0.05)
Initial weight range for all LSTM parameters.

assumptions (4)

domain assumption The Halvani/PAN corpus is a representative benchmark for forensic authorship verification on social media.
The paper's entire evaluation uses this corpus; if it is not representative, the reported gain may not transfer to real forensic cases.
domain assumption Cross-validation folds do not need to be author-disjoint for a valid accuracy estimate.
Splitting is described at instance level in Section 3.3; the paper never states that the same author is confined to one fold, and the validity of the estimate depends on the absence of identity leakage.
domain assumption GloVe embeddings plus LSTM hidden states capture enough stylistic signal for authorship discrimination.
The model relies on learned distributional representations rather than explicitly designed style features; if style information is inaccessible to these embeddings, the method could not work.
domain assumption Concatenating all known-author documents into one document preserves the relevant stylistic evidence.
The paper follows Halvani's concatenation for comparability; this choice can dilute or mix topic and style signals across documents.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Similarity Learning for Authorship Verification in Social Media." pith.science (2026). https://pith.science/paper/EQIR6KZU

@misc{pith2026190807844,
  author       = {Pith},
  title        = {Pith review of: Similarity Learning for Authorship Verification in Social Media},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EQIR6KZU}},
  note         = {Machine review of arXiv:1908.07844}
}

read the original abstract

Authorship verification tries to answer the question if two documents with unknown authors were written by the same author or not. A range of successful technical approaches has been proposed for this task, many of which are based on traditional linguistic features such as n-grams. These algorithms achieve good results for certain types of written documents like books and novels. Forensic authorship verification for social media, however, is a much more challenging task since messages tend to be relatively short, with a large variety of different genres and topics. At this point, traditional methods based on features like n-grams have had limited success. In this work, we propose a new neural network topology for similarity learning that significantly improves the performance on the author verification task with such challenging data sets.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 41 canonical work pages

[1]

A hallmark of such systems is that the true identity of anyone accessing the sys tems is typically not veriﬁed

INTRODUCTION Social media platforms and text messaging have become a perv asive way of communication in the modern world. A hallmark of such systems is that the true identity of anyone accessing the sys tems is typically not veriﬁed. As a result, users can fall victim to f alse iden- tity claims. These claims may be made for criminal purposes a nd/or the ...

work page 2019
[2]

Tradition al neural networks may learn to classify a given input

SIAMESE NETWORK TOPOLOGY Siamese networks consist of two identical neural networks w hich share the same weights as illustrated in Figure 1. Tradition al neural networks may learn to classify a given input. Siamese neural net- works, however, learn to analyze the similarity of two input s. In our case, the input to each RNN is a sequence of word embedding...

work page
[3]

EXPERIMENTAL RESULTS In this section, we present experimental results for our pro posed au- thorship veriﬁcation model with real-world social media da ta. 3.1. Dataset Currently, there are, unfortunately, not many standardized social me- dia corpora publicly available to reproduce and compare aut horship veriﬁcation results. The dataset employed in this w...

work page 2016
[4]

A modiﬁed contrasti ve loss function was chosen during system training to effectively r educe the cross-topic sensitivities of the employed word embeddi ngs

CONCLUSION We introduced a hierarchical recurrent Siamese network top ology for the task of authorship veriﬁcation. A modiﬁed contrasti ve loss function was chosen during system training to effectively r educe the cross-topic sensitivities of the employed word embeddi ngs. The proposed overall system was shown to be better adjusted than ear- lier methods ...

work page
[5]

Authorship veriﬁcat ion as a one- class classiﬁcation problem,

Moshe Koppel and Jonathan Schler, “Authorship veriﬁcat ion as a one- class classiﬁcation problem,” in Proc. ICML. 2004, pp. 62–69, ACM

work page 2004
[6]

4 ±1. 6 81. 8 ±1. 7 83. 0 ±0. 8 84. 7 ±0. 8

work page
[7]

Local histograms of character n-grams for authorship attr ibution,

Hugo Jair Escalante, Thamar Solorio, and Manuel Montes- y G´ omez, “Local histograms of character n-grams for authorship attr ibution,” in Proc. ACL. 2011, pp. 288–298, AC

work page 2011
[8]

A survey of modern authorship a ttribution methods,

Efstathios Stamatatos, “A survey of modern authorship a ttribution methods,” J. Assoc. Inf. Sci. Technol. , vol. 60, no. 3, pp. 538–556, 2009

work page 2009

Show all 47 references

[9]

Authorship attribution,

Patrick Juola, “Authorship attribution,” F oundations and Trends in Information Retrieval, vol. 1, no. 3, pp. 233–334, 2006

2006
[10]

Authorship at tribution for social media forensics,

A. Rocha, W. J. Scheirer, C. W. Forstall, T. Cavalcante, A . Theophilo, B. Shen, A. R. B. Carvalho, and E. Stamatatos, “Authorship at tribution for social media forensics,” IEEE Trans. Inf. F orensic Secur ., vol. 12, no. 1, pp. 5–33, 2017

2017
[11]

The ”fundamental problem

Moshe Koppel, Jonathan Schler, Shlomo Argamon, and Y aro n Winter, “The ”fundamental problem” of authorship attribution,” English Stud- ies, vol. 93, pp. 284–291, 2012

2012
[12]

Not all character n-grams are created equa l: A study in authorship attribution,

Upendra Sapkota, Steven Bethard, Manuel Montes y Gomez , and Thamar Solorio, “Not all character n-grams are created equa l: A study in authorship attribution,” in Proc. NAACL. ACL, 2015, pp. 93–102

2015
[13]

Authorship attrib ution using function words adjacency networks,

S. Segarra, M. Eisen, and A. Ribeiro, “Authorship attrib ution using function words adjacency networks,” in Proc. ICASSP, 2013, pp. 5563– 5567

2013
[14]

On t he usefulness of compression models for authorship veriﬁcation,

Oren Halvani, Christian Winter, and Lukas Graner, “On t he usefulness of compression models for authorship veriﬁcation,” in Proc. ARES . 2017, ACM

2017
[15]

Authorship attribution with convolutional neural networks and POS-El iding,

Julian Hitschler, Esther van den Berg, and Ines Rehbein, “Authorship attribution with convolutional neural networks and POS-El iding,” in Proc. StyleV ar. 2017, pp. 53–58, ACL

2017
[16]

C haracter-level and multi-channel convolutional neural networks for large-scale author- ship attribution,

Sebastian Ruder, Parsa Ghaffari, and John G. Breslin, “C haracter-level and multi-channel convolutional neural networks for large-scale author- ship attribution,” CoRR, vol. abs/1609.06686, 2016

2016 arXiv
[17]

Determining if two docu ments are written by the same author,

Moshe Koppel and Y aron Winter, “Determining if two docu ments are written by the same author,” Journal of the Association for Information Science and Technology, vol. 65, no. 1, pp. 178–187
[18]

Autho rship veriﬁca- tion for different languages, genres and topics,

Oren Halvani, Christian Winter, and Anika Pﬂug, “Autho rship veriﬁca- tion for different languages, genres and topics,” Digital Investigation, vol. 16, no. S, pp. S33–S43, 2016

2016
[19]

Learning a similarity metric discriminatively, with application to face veriﬁca tion,

Sumit Chopra, Raia Hadsell, and Y ann LeCun, “Learning a similarity metric discriminatively, with application to face veriﬁca tion,” in Proc. CVPR. 2005, pp. 539–546, IEEE Computer Society

2005
[20]

Text categoriza tion using compression models,

E. Frank, Chang Chui, and I. H. Witten, “Text categoriza tion using compression models,” in Proceedings DCC 2000. Data Compression Conference, 2000, pp. 555–

2000
[21]

Siamese neural networks for one-shot image recognition,

Gregory Koch, Richard Zemel, and Ruslan Salakhutdinov , “Siamese neural networks for one-shot image recognition,” 2015

2015
[22]

Part-of-speech tagging f or twitter: Annotation, features, and experiments,

Kevin Gimpel, Nathan Schneider, Brendan O’Connor, Dip anjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Y ogatama, Jef- frey Flanigan, and Noah A. Smith, “Part-of-speech tagging f or twitter: Annotation, features, and experiments,” in Proc. ACL. 2011, pp. 42–47, ACL

2011
[23]

A hierar chi- cal neural autoencoder for paragraphs and documents,

Jiwei Li, Minh-Thang Luong, and Dan Jurafsky, “A hierar chi- cal neural autoencoder for paragraphs and documents,” CoRR, vol. abs/1506.01057, 2015

2015 arXiv
[25]

Signature veriﬁcation using a

Jane Bromley, Isabelle Guyon, Y ann LeCun, Eduard S¨ ack inger, and Roopak Shah, “Signature veriﬁcation using a ”Siamese” time delay neural network,” in Proc. NIPS. 1993, pp. 737–744, Morgan Kaufmann Publishers Inc

1993
[26]

Convolutional neural networks for sentence classiﬁcation,

Y oon Kim, “Convolutional neural networks for sentence classiﬁcation,” CoRR, vol. abs/1408.5882, 2014

2014 arXiv
[27]

Discriminative deep metric l earning for face veriﬁcation in the wild,

J. Hu, J. Lu, and Y . P . Tan, “Discriminative deep metric l earning for face veriﬁcation in the wild,” in Proc. CVPR, 2014, pp. 1875–1882

2014
[28]

Overview of the author iden- tiﬁcation task at PAN 2013,

Patrick Juola and Efstathios Stamatatos, “Overview of the author iden- tiﬁcation task at PAN 2013,” in CLEF (W orking Notes), 2013, vol. 1179 of CEUR W orkshop Proceedings

2013
[29]

Deep metric learning for visua l under- standing: An overview of recent advances,

J. Lu, J. Hu, and J. Zhou, “Deep metric learning for visua l under- standing: An overview of recent advances,” IEEE Signal Processing Magazine, vol. 34, no. 6, pp. 76–84, 2017

2017
[30]

Siamese recurr ent architec- tures for learning sentence similarity,

Jonas Mueller and Aditya Thyagarajan, “Siamese recurr ent architec- tures for learning sentence similarity,” in Proc. AAAI. 2016, pp. 2786– 2792, AAAI Press

2016
[31]

L earning text similarity with Siamese recurrent networks,

Paul Neculoiu, Maarten V ersteegh, and Mihai Rotaru, “L earning text similarity with Siamese recurrent networks,” in Proc. RepL4NLP. 2016, pp. 148–157, ACL

2016
[32]

Sequenc e to sequence learning with neural networks,

Ilya Sutskever, Oriol Vinyals, and Quoc V . Le, “Sequenc e to sequence learning with neural networks,” in Proc. NIPS. 2014, pp. 3104–3112, MIT Press

2014
[33]

A theoretically grou nded appli- cation of dropout in recurrent neural networks,

Y arin Gal and Zoubin Ghahramani, “A theoretically grou nded appli- cation of dropout in recurrent neural networks,” in NIPS, D. D. Lee, M. Sugiyama, U. V . Luxburg, I. Guyon, and R. Garnett, Eds., pp. 1019–
[34]

Convolutional neural net works for authorship attribution of short texts,

Prasha Shrestha, Sebastian Sierra, Fabio Gonzalez, Ma nuel Montes, Paolo Rosso, and Thamar Solorio, “Convolutional neural net works for authorship attribution of short texts,” in Proc. EACL. 2017, pp. 669– 674, ACL

2017
[35]

Dropout: A simple way to prevent neural networks from overﬁtting,

Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov, “Dropout: A simple way to prevent neural networks from overﬁtting,” JMLR, vol. 15, pp. 1929–1958, 2014

1929
[36]

Overview of the author identiﬁc ation task at PAN 2014,

Efstathios Stamatatos, Walter Daelemans, Ben V erhoev en, Benno Stein, Martin Potthast, Patrick Juola, Miguel A. S´ anchez- P´ erez, and Alberto Barr´ on-Cede˜ no, “Overview of the author identiﬁc ation task at PAN 2014,” in CLEF (W orking Notes), 2014, vol. 1180 of CEUR W ork...

2014
[37]

Overview of the author identiﬁcation task at PAN 2015,

Efstathios Stamatatos, Walter Daelemans, Ben V erhoev en, Patrick Juola, Aurelio L ´ opez-L ´ opez, Martin Potthast, and Benno S tein, “Overview of the author identiﬁcation task at PAN 2015,” in CLEF (W orking Notes), 2015, vol. 1391 of CEUR W orkshop Proceedings

2015
[38]

Glove: Global vectors for word representation.,

Jeffrey Pennington, Richard Socher, and Christopher D Manning, “Glove: Global vectors for word representation.,” in EMNLP. 2014, pp. 1532–1543, ACM

2014
[39]

Sequence to sequence learning with neural networks,

Ilya Sutskever, Oriol Vinyals, and Quoc V Le, “Sequence to sequence learning with neural networks,” in NIPS, pp. 3104–3112. Curran Asso- ciates, Inc., 2014

2014
[40]

End-to-end sequence labe ling via Bi-directional LSTM-CNNs-CRF,

Xuezhe Ma and Eduard H. Hovy, “End-to-end sequence labe ling via Bi-directional LSTM-CNNs-CRF,” CoRR, vol. abs/1603.01354, 2016. 5

2016 arXiv
[42]

ADADELTA: an adaptive learning rat e method,

Matthew D. Zeiler, “ADADELTA: an adaptive learning rat e method,” CoRR, vol. abs/1212.5701, 2012

2012 arXiv
[44]

A large annotated corpus for learning natu ral lan- guage inference,

Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christo- pher D. Manning, “A large annotated corpus for learning natu ral lan- guage inference,” in Proc. EMNLP. 2015, ACL

2015
[45]

Semi-supervised sequence le arning,

Andrew M Dai and Quoc V Le, “Semi-supervised sequence le arning,” in Proc. NIPS. 2015, pp. 3079–3087, Curran Associates, Inc

2015
[46]

Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative ﬁ ltering,

Ruining He and Julian McAuley, “Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative ﬁ ltering,” in Proc. of the 25th International Conference on W orld Wide W eb, 2016

2016
[47]

Hierarchical attention networks for documen t classiﬁ- cation,

Zichao Y ang, Diyi Y ang, Chris Dyer, Xiaodong He, Alex Sm ola, and Eduard Hovy, “Hierarchical attention networks for documen t classiﬁ- cation,” in Proc. NAACL, 2016, pp. 1480–1489

2016
[70]

7, which is comparable to the average results reported in [11], where Table 7 shows ≈ 71

9% ± 1. 7, which is comparable to the average results reported in [11], where Table 7 shows ≈ 71. 1% accuracy for a ﬁxed test set. Comparing the results of both methods, it can readily be seen that the proposed approach signiﬁcantly outperforms the ba seline system. We were ab...
[84]

3 ±2. 1 83. 7 ±2. 1 83. 9 ±1. 1 85. 3 ±0. 9 HRSN (data augmentation + pretraining)
[1027]

Curran Associates, Inc., 2016

2016

Pith tools

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

[1] [1]

A hallmark of such systems is that the true identity of anyone accessing the sys tems is typically not veriﬁed

INTRODUCTION Social media platforms and text messaging have become a perv asive way of communication in the modern world. A hallmark of such systems is that the true identity of anyone accessing the sys tems is typically not veriﬁed. As a result, users can fall victim to f alse iden- tity claims. These claims may be made for criminal purposes a nd/or the ...

work page 2019

[2] [2]

Tradition al neural networks may learn to classify a given input

SIAMESE NETWORK TOPOLOGY Siamese networks consist of two identical neural networks w hich share the same weights as illustrated in Figure 1. Tradition al neural networks may learn to classify a given input. Siamese neural net- works, however, learn to analyze the similarity of two input s. In our case, the input to each RNN is a sequence of word embedding...

work page

[3] [3]

EXPERIMENTAL RESULTS In this section, we present experimental results for our pro posed au- thorship veriﬁcation model with real-world social media da ta. 3.1. Dataset Currently, there are, unfortunately, not many standardized social me- dia corpora publicly available to reproduce and compare aut horship veriﬁcation results. The dataset employed in this w...

work page 2016

[4] [4]

A modiﬁed contrasti ve loss function was chosen during system training to effectively r educe the cross-topic sensitivities of the employed word embeddi ngs

CONCLUSION We introduced a hierarchical recurrent Siamese network top ology for the task of authorship veriﬁcation. A modiﬁed contrasti ve loss function was chosen during system training to effectively r educe the cross-topic sensitivities of the employed word embeddi ngs. The proposed overall system was shown to be better adjusted than ear- lier methods ...

work page

[5] [5]

Authorship veriﬁcat ion as a one- class classiﬁcation problem,

Moshe Koppel and Jonathan Schler, “Authorship veriﬁcat ion as a one- class classiﬁcation problem,” in Proc. ICML. 2004, pp. 62–69, ACM

work page 2004

[6] [6]

4 ±1. 6 81. 8 ±1. 7 83. 0 ±0. 8 84. 7 ±0. 8

work page

[7] [7]

Local histograms of character n-grams for authorship attr ibution,

Hugo Jair Escalante, Thamar Solorio, and Manuel Montes- y G´ omez, “Local histograms of character n-grams for authorship attr ibution,” in Proc. ACL. 2011, pp. 288–298, AC

work page 2011

[8] [8]

A survey of modern authorship a ttribution methods,

Efstathios Stamatatos, “A survey of modern authorship a ttribution methods,” J. Assoc. Inf. Sci. Technol. , vol. 60, no. 3, pp. 538–556, 2009

work page 2009