Update README
Browse files
README.md
CHANGED
@@ -58,3 +58,45 @@ configs:
|
|
58 |
- split: test
|
59 |
path: queries/test-*
|
60 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
- split: test
|
59 |
path: queries/test-*
|
60 |
---
|
61 |
+
# ConTEB - NarrativeQA
|
62 |
+
|
63 |
+
This dataset is part of *ConTEB* (Context-aware Text Embedding Benchmark), designed for evaluating contextual embedding model capabilities. It stems from the widely used [NarrativeQA](https://huggingface.co/datasets/deepmind/narrativeqa) dataset.
|
64 |
+
|
65 |
+
## Dataset Summary
|
66 |
+
|
67 |
+
NarrativeQA (literature), consists of long documents, associated to existing sets of question-answer pairs. To build the corpus, we start from the pre-existing collection documents, extract the text, and chunk them (using [LangChain](https://github.com/langchain-ai/langchain)'s RecursiveCharacterSplitter with a threshold of 1000 characters). Since chunking is done a posteriori without considering the questions, chunks are not always self-contained and eliciting document-wide context can help build meaningful representations. We use GPT-4o to annotate which chunk, among the gold document, best contains information needed to answer the query.
|
68 |
+
|
69 |
+
This dataset provides a focused benchmark for contextualized embeddings. It includes a set of original documents, chunks stemming from them, and queries.
|
70 |
+
|
71 |
+
* **Number of Documents:** 355
|
72 |
+
* **Number of Chunks:** 1750
|
73 |
+
* **Number of Queries:** 8575
|
74 |
+
* **Average Number of Tokens per Chunk:** 151.9
|
75 |
+
|
76 |
+
## Dataset Structure (Hugging Face Datasets)
|
77 |
+
The dataset is structured into the following columns:
|
78 |
+
|
79 |
+
* **`documents`**: Contains chunk information:
|
80 |
+
* `"chunk_id"`: The ID of the chunk, of the form `doc-id_chunk-id`, where `doc-id` is the ID of the original document and `chunk-id` is the position of the chunk within that document.
|
81 |
+
* `"chunk"`: The text of the chunk
|
82 |
+
* **`queries`**: Contains query information:
|
83 |
+
* `"query"`: The text of the query.
|
84 |
+
* `"answer"`: The answer relevant to the query, from the original dataset.
|
85 |
+
* `"chunk_id"`: The ID of the chunk that the query is related to, of the form `doc-id_chunk-id`, where `doc-id` is the ID of the original document and `chunk-id` is the position of the chunk within that document.
|
86 |
+
|
87 |
+
## Usage
|
88 |
+
|
89 |
+
Use the `train` split for training, and the `test` split for evaluation.
|
90 |
+
We will upload a Quickstart evaluation snippet soon.
|
91 |
+
|
92 |
+
## Citation
|
93 |
+
|
94 |
+
We will add the corresponding citation soon.
|
95 |
+
|
96 |
+
## Acknowledgments
|
97 |
+
|
98 |
+
This work is partially supported by [ILLUIN Technology](https://www.illuin.tech/), and by a grant from ANRT France.
|
99 |
+
|
100 |
+
## Copyright
|
101 |
+
|
102 |
+
All rights are reserved to the original authors of the documents.
|