Tackling Hallucinations in Neural Chart Summarization

  Introduction

The trained model for investigations and state-of-the-art (SOTA) improvements are detailed in the paper: Tackling Hallucinations in Neural Chart Summarization. This repo contains optimized input prompts and summaries after NLI-filtering.

  Abstract

Hallucinations in text generation occur when the system produces text that is not grounded in the input. In this work, we address the problem of… See the full description on the dataset page: https://huggingface.co/datasets/saadob12/chart-to-text.

Chart Text Detection

## Overview

Chart Text Detection is a dataset for object detection tasks - it contains Text annotations for 6,399 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

There are two files in the data file, one of which is all valid comment text data used by the paper, with a total of 297,774 pieces; the other is the data required for drawing the main graphs in the paper.

Overview

This dataset covers the encoder embeddings and prediction results of LLMs of paper 'Model Generalization on Text Attribute Graphs: Principles with Lagre Language Models', Haoyu Wang, Shikun Liu, Rongzhe Wei, Pan Li.

  Dataset Description

The dataset structure should be organized as follows: /dataset/ │── [dataset_name]/ │ │── processed_data.pt # Contains labels and graph information │ │── [encoder]_x.pt # Features extracted by different encoders │… See the full description on the dataset page: https://huggingface.co/datasets/Graph-COM/Text-Attributed-Graphs.

Chart2Text is a dataset that was crawled from 23,382 freely accessible pages from statista.com in early March of 2020, yielding a total of 8,305 charts, and associated summaries. For each chart, the chart image, the underlying data table, the title, the axis labels, and a human-written summary describing the statistic was downloaded.

Currently, in the field of chart datasets, most existing resources are mainly in English, and there are almost no open-source Chinese chart datasets, which brings certain limitations to research and applications related to Chinese charts. This dataset draws on the construction method of the DVQA dataset to create a chart dataset focused on the Chinese environment. To ensure the authenticity and practicality of the dataset, we first referred to the authoritative website of the National Bureau of Statistics and selected 24 widely used data label categories in practical applications, totaling 262 specific labels. These tag categories cover multiple important areas such as socio-economic, demographic, and industrial development. In addition, in order to further enhance the diversity and practicality of the dataset, this paper sets 10 different numerical dimensions. These numerical dimensions not only provide a rich range of values, but also include multiple types of values, which can simulate various data distributions and changes that may be encountered in real application scenarios. This dataset has carefully designed various types of Chinese bar charts to cover various situations that may be encountered in practical applications. Specifically, the dataset not only includes conventional vertical and horizontal bar charts, but also introduces more challenging stacked bar charts to test the performance of the method on charts of different complexities. In addition, to further increase the diversity and practicality of the dataset, the text sets diverse attribute labels for each chart type. These attribute labels include but are not limited to whether they have data labels, whether the text is rotated 45 °, 90 °, etc. The addition of these details makes the dataset more realistic for real-world application scenarios, while also placing higher demands on data extraction methods. In addition to the charts themselves, the dataset also provides corresponding data tables and title text for each chart, which is crucial for understanding the content of the chart and verifying the accuracy of the extracted results. This dataset selects Matplotlib, the most popular and widely used data visualization library in the Python programming language, to be responsible for generating chart images required for research. Matplotlib has become the preferred tool for data scientists and researchers in data visualization tasks due to its rich features, flexible configuration options, and excellent compatibility. By utilizing the Matplotlib library, every detail of the chart can be precisely controlled, from the drawing of data points to the annotation of coordinate axes, from the addition of legends to the setting of titles, ensuring that the generated chart images not only meet the research needs, but also have high readability and attractiveness visually. The dataset consists of 58712 pairs of Chinese bar charts and corresponding data tables, divided into training, validation, and testing sets in a 7:2:1 ratio.

Dataset

This dataset was created by Ghanender Pahuja

Contents

This is the repository for ISWC 2023 Resource Track submission for Text2KGBench: Benchmark for Ontology-Driven Knowledge Graph Generation from Text. Text2KGBench is a benchmark to evaluate the capabilities of language models to generate KGs from natural language text guided by an ontology. Given an input ontology and a set of sentences, the task is to extract facts from the text while complying with the given ontology (concepts, relations, domain/range constraints) and being faithful to the input sentences.

It contains two datasets (i) Wikidata-TekGen with 10 ontologies and 13,474 sentences and (ii) DBpedia-WebNLG with 19 ontologies and 4,860 sentences.

An example

An example test sentence:

Test Sentence: {"id": "ont_music_test_n", "sent": ""The Loco-Motion" is a 1962 pop song written by American songwriters Gerry Goffin and Carole King."}

An example of ontology:

Ontology: Music Ontology

Expected Output:

{ "id": "ont_k_music_test_n", "sent": ""The Loco-Motion" is a 1962 pop song written by American songwriters Gerry Goffin and Carole King.", "triples": [ { "sub": "The Loco-Motion", "rel": "publication date", "obj": "01 January 1962" },{ "sub": "The Loco-Motion", "rel": "lyrics by", "obj": "Gerry Goffin" },{ "sub": "The Loco-Motion", "rel": "lyrics by", "obj": "Carole King" },] }

The data is released under a Creative Commons Attribution-ShareAlike 4.0 International (CC BY 4.0) License.

The structure of the repo is as the following.

Text2KGBench

src: the source code used for generation and evaluation, and baseline

benchmark the code used to generate the benchmark

evaluation evaluation scripts for calculating the results

baseline code for generating the baselines including prompts, sentence similarities, and LLM client.

data: the benchmark datasets and baseline data. There are two datasets: wikidata_tekgen and dbpedia_webnlg.

wikidata_tekgen Wikidata-TekGen Dataset

ontologies 10 ontologies used by this dataset

train training data

test test data

manually_verified_sentences ids of a subset of test cases manually validated

unseen_sentences new sentences that are added by the authors which are not part of Wikipedia

test unseen test unseen test sentences

ground_truth ground truth for unseen test sentences.

ground_truth ground truth for the test data

baselines data related to running the baselines.

test_train_sent_similarity for each test case, 5 most similar train sentences generated using SBERT T5-XXL model.

prompts prompts corresponding to each test file

unseen prompts unseen prompts for the unseen test cases

Alpaca-LoRA-13B data related to the Alpaca-LoRA model

llm_responses raw LLM responses and extracted triples

eval_metrics ontology-level and aggregated evaluation results

unseen results results for the unseen test cases

llm_responses raw LLM responses and extracted triples

eval_metrics ontology-level and aggregated evaluation results

Vicuna-13B data related to the Vicuna-13B model

llm_responses raw LLM responses and extracted triples

eval_metrics ontology-level and aggregated evaluation results

dbpedia_webnlg DBpedia Dataset

ontologies 19 ontologies used by this dataset

train training data

test test data

ground_truth ground truth for the test data

baselines data related to running the baselines.

test_train_sent_similarity for each test case, 5 most similar train sentences generated using SBERT T5-XXL model.

prompts prompts corresponding to each test file

Alpaca-LoRA-13B data related to the Alpaca-LoRA model

llm_responses raw LLM responses and extracted triples

eval_metrics ontology-level and aggregated evaluation results

Vicuna-13B data related to the Vicuna-13B model

llm_responses raw LLM responses and extracted triples

eval_metrics ontology-level and aggregated evaluation results

This benchmark contains data derived from the TekGen corpus (part of the KELM corpus) [1] released under CC BY-SA 2.0 license and WebNLG 3.0 corpus [2] released under CC BY-NC-SA 4.0 license.

[1] Oshin Agarwal, Heming Ge, Siamak Shakeri, and Rami Al-Rfou. 2021. Knowledge Graph Based Synthetic Corpus Generation for Knowledge-Enhanced Language Model Pre-training. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3554–3565, Online. Association for Computational Linguistics.

[2] Claire Gardent, Anastasia Shimorina, Shashi Narayan, and Laura Perez-Beltrachini. 2017. Creating Training Corpora for NLG Micro-Planners. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 179–188, Vancouver, Canada. Association for Computational Linguistics.

Open Text reported $12.76B in Market Capitalization this December of 2025, considering the latest stock price and the number of outstanding shares.Data for Open Text | OTC - Market Capitalization including historical, tables and charts were last updated by Trading Economics this last December in 2025.

"Scholarly figures are data visualizations like bar charts, pie charts, line graphs, maps, scatter plots or similar figures. Text extraction from scholarly figures is useful in many application scenarios, since text in scholarly figures often contains information that is not present in the surrounding text. This dataset is a corpus of 121 scholarly figures from the economics domain evaluating text extraction tools. We randomly extracted these figures from a corpus of 288,000 open access publications from EconBiz. The dataset resembles a wide variety of scholarly figures from bar charts to maps. We manually labeled the figures to create the gold standard.

We adjusted the provided gold standard to have a uniform format for all datasets. Each figure is accompanied by a TSV file (tab-separated values) where each entry corresponds to a text line which has the following structure:

X-coordinate of the center of the bounding box in pixel

Y-coordinate of the center of the bounding box in pixel

Width of the bounding box in pixel

Height of the bounding box in pixel

Rotation angle around its center in degree

Text inside the bounding box

In addition we provide the ground truth in JSON format. A schema file is included in each dataset as well. The dataset is accompanied with a ReadMe file with further information about the figures and their origin.

If you use this dataset in your own work, please cite one of the papers in the references."

Number of words and number of tokens denote the number of words in the dataset before and after preprocessing respectively. Direct edges and indirect edges represent the number of direct and indirect synonym relationships between words in the text graph respectively.

Open Text reported $13.48B in Assets for its fiscal quarter ending in September of 2025. Data for Open Text | OTC - Assets including historical, tables and charts were last updated by Trading Economics this last December in 2025.

We provide a new, publicly available dataset of the extracted knowledge graphs (from both REBEL and Gemini) for the Reuters-21578, BBC and AG News and 20 news groups benchmarks. This resource can be used to benchmark other graph-based and knowledge-aware classification methods.

This is the dataset of our EMNLP 2021 paper:

Graphine: A Dataset for Graph-aware Terminology Definition Generation.

Please read the "readme.md" in it for the format of the dataset.

The data this week comes from Data.World by way of Sean Miller, Billboard.com and Spotify.

Billboard Top 100 - Wikipedia

The Billboard Hot 100 is the music industry standard record chart in the United States for songs, published weekly by Billboard magazine. Chart rankings are based on sales (physical and digital), radio play, and online streaming in the United States.

Billboard Top 100 Article

Drake rewrites the record for the most entries ever on the Billboard Hot 100, as he lands his 208th career title on the latest list, dated March 21

Data Dictionary

billboard.csv

audio_features.csv

Graph database has developed rapidly and plays an important role in research nowadays. It helps scientists in various ways, e.g., finding related works, exploring works in a research area, or gaining knowledge from connections between different nodes. There are already some graph databases for research available on the Internet. However, they do not meet the needs of Digital Humanities (DH) scientists, who mainly work with historical data. Therefore, we create a graph database specifically for DH scientists. This database is part of MINE, a service that facilitates data acquisition and big data analysis.

Number of words and number of tokens denote the number of words in the dataset before and after preprocessing respectively. Direct edges and indirect edges represent the number of direct and indirect synonym relationships between words in the text graph respectively.

Open Text reported $4.57M in Debt for its fiscal quarter ending in June of 2025. Data for Open Text | OTC - Debt including historical, tables and charts were last updated by Trading Economics this last December in 2025.

and semantic bias

EduMKG is a multimodal educational knowledge graph dataset that covers natural sciences (biology, physics, and chemistry) in middle and high school education. It includes multimodal concepts covering text, images, videos, and audio, as well as knowledge points and exercises extracted from curriculum standards and MOOCs. EduMKG comprises 34,630 multimodal concepts and 403,400 triples, making it a vital resource for research in multimodal educational applications.

1. We have open-sourced the RDF data of EduMKG (following IRI standards)
2. We have open-sourced an automation script for converting JSON to RDF format on GitHub at https://github.com/AI-BNU-TEAMKG/EduMKG.
3. We have released a SPARQL endpoint and provided basic usage examples.
4. We have released the schema of EduMKG on GitHub at https://github.com/AI-BNU-TEAMKG/EduMKG.
5. We have released the validation results for alignment quality on GitHub at https://github.com/AI-BNU-TEAMKG/EduMKG

New Document Description

EduMKG.rdf : The RDF-formatted knowledge graph adheres to the IRI standards and has already been uploaded to this dataset.

Accessing the SPARQL Endpoint and Performing Queries: We provide an example for reference.

• Access the Apache Jena Fuseki UI

• Enter the username and password: user, userPassword.

• Example usage:

  Example 1

  
  # Query the concept of “上臂骨骼肌” corresponding to its explanation
  PREFIX ex: <http://v1.edumkg.org/>
  SELECT ?explanation
  WHERE {
   ?concept a ex:Concept .
   ?concept ex:hasAnExplanation ?explanation .
   FILTER(CONTAINS(STR(?concept), ENCODE_FOR_URI("上臂骨骼肌")))
  }

  Example 2

  # Randomly select 10 sets of "subject-predicate-object" triplet data from the database and display them.
  PREFIX ex: <http://v1.edumkg.org/
  SELECT ?subject ?predicate ?object
  WHERE {
  ?subject ?predicate ?object
  }
  LIMIT 10

Zip Description
RawData: This folder comprises the publicly available raw data, including images derived from course materials, subtitles extracted from MOOC videos, audio files associated with specific concepts, and supplementary files generated through multimodal concept extraction and alignment.
EduMKG: This folder contains the entire EduMKG, including the triple files of EduMKG as well as the cross-indexing files between multimodal concepts, knowledge points, and images
Example: The examples of the Case Study for Multimodal Alignment Section in the paper.
Groundtruth: The groundtruth for Evaluation on Concept Extraction Section in the paper.

Document Description

Documents in EduMKG
{subject}Image.json: This is a cross-indexing file between images and concepts in the multimodal discipline. Each image has a unique ID and is associated with a varying number of concepts.
{subject}Knowledge.json: This is a cross-reference file for knowledge points. Each knowledge point has a unique ID, along with URLs for corresponding exercises and IDs of related knowledge points.
{subject}.json: This is a multimodal concept cross-indexing file involving text, images, videos, audio, and knowledge points. Each multimodal concept has a unique ID.
{subjectTriples}.json: This file contains all the triples of EduMKG.

Documents in Example

Each folder contains text, image, video, and audio knowledge of multimodal concepts. To avoid copyright issues, we only provide video URLs and the time intervals where the concepts appear.

Documents in Groundtruth

Each `groundtruth.txt` contains the concepts corresponding to the course with the same name as the folder. The contents were manually annotated by three PhD students and seven master's students, with cross-validation performed to ensure accuracy. The `groundtruth.txt` files are used in the "Evaluation on Concept Extraction" section of the paper to validate the effectiveness of the proposed concept extraction method.

Documents in RawData
Includes images corresponding to multimodal concepts, as well as intermediate files generated during the construction of the knowledge graph, facilitating the reproducibility of the EduMKG construction process.