Ali Abusaleh

Research Assistant

Portrait of Ali Abusaleh

Contact

Goethe-Universität Frankfurt am Main
Robert-Mayer-Straße 10
Room 401e
D-60325 Frankfurt am Main
D-60629 Frankfurt am Main (use for package delivery)
Postfach / P.O. Box: 154
Phone:
Mail:

Google Scholar · ORCID · GitHub · LinkedIn

Project

New Data Spaces for the Social Sciences (SPP 2431)

ENTAILab – CIRCLET: research-driven infrastructure for survey-related text, image, audio and video data. Funded by DFG (SPP 2431).

Research

My research focuses on large language models and retrieval-augmented generation, multimodal AI and vision-language models, privacy-preserving NLP and Arabic NLP. In ENTAILab, I develop DUUI components for CIRCLET, including multimodal model serving and the anonymization of personal data in research texts.

Thesis topic proposals

2026

Master Thesis: Generative Frameworks for Multimodal Survey Data Anonymization.
Description
Ensuring privacy in open-ended survey research demands sophisticated anonymization frameworks capable of sanitizing text, images, and audio without destroying the downstream research utility of the data. This Master's thesis explores the intersection of deep generative modeling and privacy preservation to build an end-to-end multimodal anonymization pipeline. For text, the student will investigate advanced pseudonymization and contextual text-replacement strategies leveraging state-of-the-art LLMs and privacy-filtering guardrails. For visual data, the project will combine semantic segmentation and OCR with generative Diffusion Models to perform secure, context-aware inpainting (e.g., replacing faces or identifying text while preserving the overall scene context). For audio, the student will explore voice-masking and speech synthesis techniques to redact speaker identity. The core research focus will be a rigorous, quantitative evaluation of the trade-off between strict privacy preservation and the semantic utility of the sanitized datasets for downstream empirical analysis.
Corresponding Lab Member: Ali Abusaleh and Alexander Mehler.
Bachelor Thesis: Joint Sarcasm, Sentiment, and Stance Detection in Underrepresented Dialects.
Description
Colloquial communication on social media frequently blends emotional expression with social or political viewpoints, making the simultaneous detection of sentiment, stance, and sarcasm highly interconnected. Sarcasm heavily co-exists with these features, often acting as a linguistic disruptor that flips sentiment polarity and obscures an author's true stance toward a target. This thesis involves developing a unified prompt-based or multi-task classification pipeline to jointly model sarcasm, sentiment, and stance in dialectal text. The research will primarily anchor on Arabic dialects (such as Levantine, Egyptian, and Gulf) and extend to underrepresented Indic dialects (such as regional variants of Hindi or Bhojpuri). The student will leverage state-of-the-art dialectal and multilingual language models (e.g., MARBERTv2, IndicBERT) to evaluate how models capture subjective alignment under low-resource constraints. Tasks include benchmarking multi-dialectal datasets, implementing joint classification layers, and analyzing error patterns caused by sarcasm-induced stance flips.
Corresponding Lab Member: Ali Abusaleh and Alexander Mehler.

2025

Bachelor Thesis: Multimodal Sentiment Analysis via Cross-Attention Fusion in Latent Space.
Description
Accurately estimating sentiments in video data is a complex challenge that requires integrating visual, audio, and textual cues. This proposal involves developing a multimodal sentiment analysis pipeline that uses state-of-the-art models to extract embeddings from different modalities: Qwen2.5 for text, JEPA 2 for visuals, and WhisperX (pyannote) for audio. The embeddings then undergo projection into a shared latent space and fusion via cross-attention mechanisms to capture intermodal dependencies. A probabilistic output layer will then estimate sentiment distributions over fine-grained emotions. To accelerate the analysis process for real-time applications, the pipeline should use a stream-like vector projection method that updates latent space representations incrementally instead of reprocessing entire sequences. The pipeline will be built using DUUI to ensure modularity, scalability, and reproducibility. The goal of this thesis is to tackle limitations of unimodal sentiment analysis and traditional fusion methods to achieve efficient, accurate, and scalable multimodal sentiment analysis.
Corresponding Lab Member: Ali Abusaleh and Alexander Mehler.

If you have any suggestions of your own relating to this or our other proposed topics, please do not hesitate to contact us.

In addition, we provide a mailing list for free, which we use to inform regularly about updates on new qualification and research work as well as other information relating to Texttechnology.


Publications

2026

Ali Abusaleh, Giuseppe Abrami, Mevlüt Bagci, Patrick Schrottenbacher, Alexander Mehler, J. Weiß, G. Gühring and J. Gorges. November, 2026. Anonymisation Infrastructure for Social Science Research Data: Landscape, Challenges, and Directions.
BibTeX
@misc{abusaleh-etal-2026-anonymisation,
  title     = {Anonymisation Infrastructure for Social Science Research Data:
               Landscape, Challenges, and Directions},
  author    = {Abusaleh, Ali and Abrami, Giuseppe and Bagci, Mevl{\"u}t and Schrottenbacher, Patrick
               and Mehler, Alexander and Wei{\ss}, J. and G{\"u}hring, G. and Gorges, J.},
  howpublished = {New Data Spaces | Reports, No.~3. DFG Infrastructure Priority
                  Programme ``New Data Spaces for the Social Sciences'' (SPP 2431)},
  month     = {nov},
  year      = {2026},
  doi       = {10.5157/SPP2431:WP3:1.0},
  url       = {https://www.new-data-spaces.de/en-us/Start/Publications/New-Data-Spaces-Reports/Detail/detail/wp-3},
  abstract  = {The anonymisation of research data in the social sciences remains
               a persistent bottleneck to open science and cross-institutional
               collaboration. Despite decades of research on de-identification,
               deployed systems continue to struggle with quasi-identifiers,
               context preservation, third-party mentions, and multimodal content,
               while the legal framework governing pseudonymisation became more
               rather than less ambiguous in 2025 following contradictory guidance
               from the European Data Protection Board and the Court of Justice
               of the European Union. This report examines the technical and
               regulatory landscape of anonymisation for social science research
               data and identifies open challenges through a literature review
               and a small-scale empirical evaluation of three current personally
               identifiable information (PII) detection backends, surfacing systematic
               issues in tokenisation, label-scheme consistency, and coverage
               of GDPR Article 9 special categories. Drawing on the concrete
               needs voiced by research projects working with survey, interview,
               archival, clinical, and chat data, the paper outlines directions
               for shared, reusable anonymisation infrastructure in the social
               sciences, and presents DUUI-Anonymize as one worked example: a
               component that lets projects test existing PII detection backends
               through a common interface. The paper builds on discussions from
               the CIRCLET Anonymisation Workshop, held in July 2026 within the
               DFG-funded infrastructure priority programme ``New Data Spaces
               for the Social Sciences'' (SPP 2431).}
}
Bhuvanesh Verma, Ali Abusaleh and Alexander Mehler. October, 2026. TTLab at Daleel 2026: STAR-Ar, Sequence Tagging for Argument Recognition in Arabic. Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, 1304–1309.
BibTeX
@inproceedings{verma-etal-2026-ttlab-daleel,
  title     = {{TTL}ab at Daleel 2026: {STAR}-Ar, Sequence Tagging for Argument
               Recognition in {A}rabic},
  author    = {Verma, Bhuvanesh and Abusaleh, Ali and Mehler, Alexander},
  editor    = {Abdul-Mageed, Muhammad and Elsayed, Tamer and Bouamor, Houda and El-Beltagy, Samhaa
               and Al-Khalifa, Hend and Abdelali, Ahmed and Elmadany, AbdelRahim
               and El Mekki, Abdellah and Darwish, Kareem and Alkhereyf, Sakhar and Luqman, Hamzah
               and Khalifa, Salam and Magdy, Samar and Alyafeai, Zaid and Hamed, Injy
               and Zaghouani, Wajdi and Zitouni, Imed},
  booktitle = {Proceedings of The Fourth {A}rabic Natural Language Processing
               Conference: Shared Tasks},
  month     = {oct},
  year      = {2026},
  address   = {Budapest, Hungary},
  publisher = {Association for Computational Linguistics},
  url       = {https://aclanthology.org/2026.arabicnlp-sharedtasks.163/},
  pages     = {1304--1309},
  isbn      = {979-8-89176-497-2},
  abstract  = {Argument Mining (AM) is a critical NLP task that remains significantly
               under-resourced in Arabic. This paper presents STAR-Ar, a BERT-BiLSTM-CRF
               architecture for argument discourse detection and classification,
               as our system for Daleel 2026, the inaugural Arabic argument mining
               shared task. The task requires the identification and classification
               of argumentative discourse units (ADUs) in debate and editorial
               texts. We jointly model these two objectives as a token-level
               sequence labeling task using a BERT-BiLSTM-CRF architecture that
               combines contextual transformer embeddings with structural transition
               constraints to support accurate span detection. STAR-Ar achieves
               an F1-score of 72.69 on validation and 73.7 on test data. Our
               domain-specific analysis shows that models trained exclusively
               on editorials underperform those trained on debates, a disparity
               we primarily attribute to the smaller size of the editorial dataset.}
}
Ali Abusaleh, Bhuvanesh Verma and Alexander Mehler. October, 2026. TTLab at AlexandriaX-2026: A Fine-Tuned Surface Tagger for Arabic Machine-Translation Error-Span Detection and Classification. Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, 786–792.
BibTeX
@inproceedings{Abusaleh:et:al:2026:mt,
  title     = {{TTL}ab at {A}lexandria{X}-2026: A Fine-Tuned Surface Tagger for
               {A}rabic Machine-Translation Error-Span Detection and Classification},
  author    = {Abusaleh, Ali and Verma, Bhuvanesh and Mehler, Alexander},
  editor    = {Abdul-Mageed, Muhammad and Elsayed, Tamer and Bouamor, Houda and El-Beltagy, Samhaa
               and Al-Khalifa, Hend and Abdelali, Ahmed and Elmadany, AbdelRahim
               and El Mekki, Abdellah and Darwish, Kareem and Alkhereyf, Sakhar and Luqman, Hamzah
               and Khalifa, Salam and Magdy, Samar and Alyafeai, Zaid and Hamed, Injy
               and Zaghouani, Wajdi and Zitouni, Imed},
  booktitle = {Proceedings of The Fourth {A}rabic Natural Language Processing
               Conference: Shared Tasks},
  month     = {oct},
  year      = {2026},
  address   = {Budapest, Hungary},
  publisher = {Association for Computational Linguistics},
  url       = {https://aclanthology.org/2026.arabicnlp-sharedtasks.98/},
  pages     = {786--792},
  isbn      = {979-8-89176-497-2},
  abstract  = {We present TTLab{'}s submission to the AlexandriaX-2026 Subtask
               3 on Arabic MT error span detection and classification. Our system
               frames the task as token-level classification over surface forms,
               preserving character offsets to ensure exact alignment with the
               evaluation metric. To handle severe label imbalance, we employ
               a focal loss with class weighting and dialect-specific decoding
               thresholds. Among six Arabic pre-trained encoders, MARBERTv2 achieves
               the best overall performance of 40.8 and 40.91 on the development
               and test set, respectively, ranking 3rd out of all participating
               teams. While our system localizes error spans effectively, classification
               of rare error types remains challenging, highlighting the need
               for data augmentation for tail categories.}
}
Bhuvanesh Verma, Ali Abusaleh and Alexander Mehler. October, 2026. TTLab at StanceEval-2026: A Cloze-Style Prompting Approach for Arabic-Language Stance Detection (CLASP-Ar). Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, 511–518.
BibTeX
@inproceedings{Verma:et:al:2026:CLASP,
  title     = {{TTL}ab at {S}tance{E}val-2026: A Cloze-Style Prompting Approach
               for {A}rabic-Language Stance Detection ({CLASP}-Ar)},
  author    = {Verma, Bhuvanesh and Abusaleh, Ali and Mehler, Alexander},
  editor    = {Abdul-Mageed, Muhammad and Elsayed, Tamer and Bouamor, Houda and El-Beltagy, Samhaa
               and Al-Khalifa, Hend and Abdelali, Ahmed and Elmadany, AbdelRahim
               and El Mekki, Abdellah and Darwish, Kareem and Alkhereyf, Sakhar and Luqman, Hamzah
               and Khalifa, Salam and Magdy, Samar and Alyafeai, Zaid and Hamed, Injy
               and Zaghouani, Wajdi and Zitouni, Imed},
  booktitle = {Proceedings of The Fourth {A}rabic Natural Language Processing
               Conference: Shared Tasks},
  month     = {oct},
  year      = {2026},
  address   = {Budapest, Hungary},
  publisher = {Association for Computational Linguistics},
  url       = {https://aclanthology.org/2026.arabicnlp-sharedtasks.66/},
  pages     = {511--518},
  isbn      = {979-8-89176-497-2},
  abstract  = {Arabic-language stance detection remains challenging, and previous
               shared-task systems have largely relied on multitask learning
               and ensembles. While these systems achieve state-of-the-art performance,
               their applicability and transferability are limited by the additional
               complexity introduced by multitask learning. To reduce this complexity,
               we introduce $\\texttt{CLASP-Ar}$, which reformulates the task
               as cloze-style masked language modeling. In this approach, the
               target, predicted sentiment, and text are combined into a single
               prompt whose $\\texttt{[MASK]}$ prediction is restricted to a
               verbalizer-constrained label vocabulary. When evaluated on the
               StanceEval-2026: Arabic Stance Detection Shared Task, our proposed
               approach achieves an $F_{\\text{avg2}}$ score of 71.36{\\%} on
               Track 1 and 74.14{\\%} on Track 2 test sets.}
}
Mevlüt Bagci, Ali Abusaleh, Daniel Baumartz, Alexander Mehler, Giuseppe Abrami and Maxim Konca. 2026. Extending a Parliamentary Corpus with MPs’ Tweets: Automatic Annotation and Evaluation Using TTLABTWEETCORPUS. KONVENS 2026 - Context matters: NLP beyond Text. accepted.
BibTeX
@inproceedings{bagci:et:al:2026,
  title     = {Extending a Parliamentary Corpus with {MP}s{\textquoteright} Tweets:
               Automatic Annotation and Evaluation Using {TTLABTWEETCORPUS}},
  author    = {Mevlüt Bagci and Ali Abusaleh and Daniel Baumartz and Alexander Mehler
               and Giuseppe Abrami and Maxim Konca},
  booktitle = {KONVENS 2026 - Context matters: NLP beyond Text},
  year      = {2026},
  address   = {Hamburg (Germany)},
  keywords  = {Corpora, Tweets, Political Data, Text, Media classification},
  note      = {accepted}
}
Christoph Wigbels, Ali Abusaleh, Markus T. Jansen, Alexander Mehler, Manuel Schaaf and Markus J. Hofmann. 2026. Individual Text Corpora Predict User-Specific Knowledge: Benchmarks of Individualized Knowledge Simulation. KONVENS 2026 - Context matters: NLP beyond Text. accepted.
BibTeX
@inproceedings{Wigbels:et:al:2026:a,
  author    = {{Wigbels, Christoph and Abusaleh, Ali} and Jansen, Markus T. and Mehler, Alexander
               and Schaaf, Manuel and Hofmann, Markus J.},
  title     = {Individual Text Corpora Predict User-Specific Knowledge: Benchmarks
               of Individualized Knowledge Simulation},
  booktitle = {KONVENS 2026 - Context matters: NLP beyond Text},
  year      = {2026},
  address   = {Hamburg (Germany)},
  keywords  = {Individual Text Corpora, Retrieval-Augmented Generation, Personalized Language Models, Probabilistic Calibration, Knowledge Benchmarks, Data Contamination, German NLP, spp, circlet},
  note      = {accepted}
}
Julia Weiss, Axel Burger, Joss Roßmann, Jan Eric Meurer and Ali Abusaleh. May, 2026. From Images to Topics: Evaluating Vision-Language Models for Topic Classification of Election Advertising. Companion Publication of the 2026 18th ACM Web Science Conference, 10–14.
BibTeX
@inproceedings{weiss:et:al:2026,
  series    = {WebSci Companion ’26},
  title     = {From Images to Topics: Evaluating Vision-Language Models for Topic
               Classification of Election Advertising},
  url       = {http://dx.doi.org/10.1145/3795513.3807426},
  doi       = {10.1145/3795513.3807426},
  booktitle = {Companion Publication of the 2026 18th ACM Web Science Conference},
  publisher = {ACM},
  author    = {Weiss, Julia and Burger, Axel and Roßmann, Joss and Meurer, Jan Eric
               and Abusaleh, Ali},
  year      = {2026},
  month     = {May},
  pages     = {10–14},
  collection = {WebSci Companion ’26}
  keywords  = {Multimodal Large Language Models, Political communication, Privacy-aware AI, new-data-spaces, circlet}
}
Ali Abusaleh, Bhuvanesh Verma and Alexander Mehler. May, 2026. TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based Fusion for Multi-Dialectal Arabic. The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks, 262–268.
BibTeX
@inproceedings{Abusaleh:et:al:2026:sarf,
  title     = {TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based
               Fusion for Multi-Dialectal Arabic},
  author    = {Abusaleh, Ali and Verma, Bhuvanesh and Mehler, Alexander},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing
               Tools (OSACT7) with 5 Shared Tasks},
  month     = {May},
  year      = {2026},
  pages     = {262--268},
  address   = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor    = {Al-Khalifa, Hend and El-Haj, Mo and Ezzini, Saad},
  doi       = {10.63317/4wj6s3ys5osk},
  keywords  = {NLP, Sentiment Analysis, Arabic analysis, new-data-spaces, circlet, satek},
  abstract  = {Arabic sentiment analysis is challenged by morphological complexity
               and lexical variation across Arabic dialects, compounded by subjectivity
               in how speakers and writers express sentiment. In this paper,
               we present our submission for the AraSentEval 2026 Shared Task
               on Arabic Dialect Sentiment Analysis. We propose SARF (صرف) a
               multi-view architectural framework that integrates surface-level
               context with stemmed and rooted morphological perspectives using
               a shared MARBERTv2 encoder. Our system employs a hybrid BERT-CNN-BiLSTM-Attention
               architecture to capture both local sentiment n-grams and global
               sequential dependencies. Experimental results show that while
               individual morphological normalization strategies (stemming or
               rooting) may degrade performance, their joint integration via
               cross-morphological attention provides robust features across
               diverse dialects. Our final system achieved a competitive macro-F1-score
               of 0.9263, ranking 2nd out of 15 participating teams.}
}
Ali AbuSaleh, Leon Hammerla and Alexander Mehler. 2026. Learning to Detect Cross-Modal Negation: An Analysis of Latent Representations and an Attention-Based Solution. 2026 8th International Conference on Natural Language Processing (ICNLP), 613–622.
BibTeX
@inproceedings{Abusaleh:et:al:2026,
  author    = {AbuSaleh, Ali and Hammerla, Leon and Mehler, Alexander},
  booktitle = {2026 8th International Conference on Natural Language Processing (ICNLP)},
  title     = {Learning to Detect Cross-Modal Negation: An Analysis of Latent
               Representations and an Attention-Based Solution},
  year      = {2026},
  volume    = {},
  number    = {},
  pages     = {613-622},
  keywords  = {Modeling;Videos;Labeling;Visualization;Signal detection;Large language models;Head;Media;Accuracy;Annotations;Vision language model;Natural language processing;Cross-modal retrieval;negation detection;video analysis;Multimodal analysis;Political Communication, neglab, new-data-spaces, circlet},
  doi       = {10.1109/ICNLP69856.2026.11527861},
  abstract  = {Detecting high-level semantic concepts like negation across modalities
               remains a challenge for current multimodal systems. We analyze
               this as a fundamental representation learning problem, providing
               the first evidence that negation does not form a linearly or non-linearly
               separable class in the latent spaces of standard vision-language
               models (VLMs). We demonstrate that pretrained embeddings primarily
               encode modality-specific features, lacking a generalizable negation
               signal. To overcome this, we propose a novel cross-modal attention
               architecture that explicitly models inter-modal dependencies,
               achieving performance gains of up to +7.03% F1 over unimodal baselines.
               Our analysis reveals a key asymmetry: while textual negation often
               appears independently, visual negation is semantically dependent
               on linguistic context, a finding validated through our statistical
               analysis of 3,222 political video-text pairs automatically annotated
               via Qwen2.5-VL. By combining this analysis with self-supervised
               video representations (JEPA2), we advance the modeling of temporal
               negation. This work provides new methods and insights for learning
               robust, semantically-aligned representations in multimodal systems.}
}

2025

Mehdi Rahim and Ali Abusaleh. 2025. GENERATIVE AI ON CGM: TOWARDS A FOUNDATION MODEL FOR GLUCOSE PREDICTION, ROOT CAUSE ANALYSIS AND ANOMALY DETECTION. DIABETES TECHNOLOGY & THERAPEUTICS, 27:E144–E144.
BibTeX
@inproceedings{rahim2025generative,
  title     = {GENERATIVE AI ON CGM: TOWARDS A FOUNDATION MODEL FOR GLUCOSE PREDICTION,
               ROOT CAUSE ANALYSIS AND ANOMALY DETECTION},
  author    = {Rahim, Mehdi and Abusaleh, Ali},
  booktitle = {DIABETES TECHNOLOGY \& THERAPEUTICS},
  volume    = {27},
  pages     = {E144--E144},
  year      = {2025},
  organization = {MARY ANN LIEBERT, INC 140 HUGUENOT STREET, 3RD FL, NEW ROCHELLE, NY 10801 USA}
}

2024

Ali Abusaleh and Mehdi Rahim. 2024. A Multitask VAE for Time Series Preprocessing and Prediction of Blood Glucose Level.
BibTeX
@misc{Abusaleh:Rahim:2024,
  title     = {A Multitask VAE for Time Series Preprocessing and Prediction of
               Blood Glucose Level},
  author    = {Ali Abusaleh and Mehdi Rahim},
  year      = {2024},
  eprint    = {2410.00015},
  archiveprefix = {arXiv},
  primaryclass = {eess.SP},
  url       = {https://arxiv.org/abs/2410.00015}
}