First, clone this repository:
git clone https://github.com/lovodkin93/highlights-extract-app.git
cd highlights-extract-app
then, run:
npm install
npm i react
to install all the packages used in this project and serve (which will be used to build the project for production).
To start working with the UI, run:
npm start
This will run the UI in the development mode.
Open http:https://localhost:3000 to view it in your browser.
The page will reload when you make changes.
You may also see any lint errors in the console.
In order to see specific datapoint to annotate, you will first need to generate a data_for_mturk.json file and save it under the data folder.
To do that, follow the following steps:
0. install the conda environment controlled_reduction and activate it:
conda env create -f generate_input_json_file/controlled_reduction.yml
conda activate controlled_reduction
- run
python3 generate_input_json_file\get_coref_clusters_chr.py -i <DATA_PATH> -o <SPACY_TOKENIZATION_OUTPUT_PATH>
<DATA_PATH> should contain the following structure where a summary and its related document directory share the same name:
- <DATA_PATH>
- summaries
- A.txt
- B.txt
- ...
- A
- doc_A1
- doc_A2
- ...
- B
- doc_B1
- doc_B2
- ...
- run
python3 generate_input_json_file\create_json_for_mturk.py -i <DATA_PATH> -o <JSON_FOR_MTURK_OUTPUT_PATH> --coref-cluster-json /path/to/coref_clusters.json
<DATA_PATH> is the same as before, and /path/to/coref_clusters.json should be <SPACY_TOKENIZATION_OUTPUT_PATH>/coref_clusters.json (the output of the code in step 1)
Finally, in <JSON_FOR_MTURK_OUTPUT_PATH> you will find the data_for_mturk.json which should be placed in the data folder.
Once data_for_mturk.json was generated and placed in the data folder, to annotate a specific datapoint:
- activate the app:
npm start
- when the url is activated, add to it
id=<DATAPOINT_ID>parameter, where<DATAPOINT_ID>is the serial index of the datapoint in the<DATA_PATH>folder (beginning with 0). For example:
http:https://localhost:3000/?id=0
would load the first sommar-document pair in <DATA_PATH> to annotate.
The app is designed to be incorporated into the mturl platform. To do that:
- generate the relevant
data_for_mturk.json - build the app, by running:
npm run build
This will build the application and save it in a build folder.
3. open a web-hosting platform (e.g. platformanywhere) and copy there the following folders: backend and build, as well as the files package.json and package-lock.json
Now, you can upload to mturk the hits using the url of the web-hosting platform, combined with the ids of your document-summary pairs. In other words, each combination of the url of the web-hosting platform with an id parameter would be a single hit. For example:
https://<USER_NAME>.pythonanywhere.com/?id=0
would upload the first document-summary pair.
To build the UI for production to the build folder, run:
npm run build
Then, to load the UI to port 8000 (for example), run:
serve -s build -p 8000
If you find this repository useful in your research, please cite the following paper:
@inproceedings{slobodkin-etal-2022-controlled,
title = "Controlled Text Reduction",
author = "Slobodkin, Aviv and
Roit, Paul and
Hirsch, Eran and
Ernst, Ori and
Dagan, Ido",
booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2022",
address = "Abu Dhabi, United Arab Emirates",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.emnlp-main.385",
pages = "5699--5715",
abstract = "Producing a reduced version of a source text, as in generic or focused summarization, inherently involves two distinct subtasks: deciding on targeted content and generating a coherent text conveying it. While some popular approaches address summarization as a single end-to-end task, prominent works support decomposed modeling for individual subtasks. Further, semi-automated text reduction is also very appealing, where users may identify targeted content while models would generate a corresponding coherent summary.In this paper, we focus on the second subtask, of generating coherent text given pre-selected content. Concretely, we formalize \textit{Controlled Text Reduction} as a standalone task, whose input is a source text with marked spans of targeted content ({``}highlighting{''}).A model then needs to generate a coherent text that includes all and only the target information.We advocate the potential of such models, both for modular fully-automatic summarization, as well as for semi-automated human-in-the-loop use cases.Facilitating proper research, we crowdsource high-quality dev and test datasets for the task. Further, we automatically generate a larger {``}silver{''} training dataset from available summarization benchmarks, leveraging a pretrained summary-source alignment model.Finally, employing these datasets, we present a supervised baseline model, showing promising results and insightful analyses.",
}