Explore projects

Stage de Master 1 de Laura Clain

encadrants: Benoît Simon Bouhet et Matthieu Authier

A Python package to process and represent events from ontologies and semi-structured databases such as Wikidata and Wikipedia.

Requests to collect documents relating real-world events (themselves described using wikivents) stored in a global index (provided by database_infrastructure_text_mining).

Textual Search Engine Infrastructure based on ElasticSearch (https://www.elastic.co/fr/elasticsearch/) and Lucene (https://lucene.apache.org/). Includes the import scripts to load datasets into the index.

Command Line Tools to manipulate the document_tracking architecture. It allows to train the Miranda algorithm, to use it and the alternative one, the K-Means implementation. It also provides a tool to evaluate the results.

Resources and Python API to manipulate datasets of news documents. It manipulates data in the .pickle format with the help of pandas and numpy. It can perform operations on the datasets.

Implementation of algorithms to detect and track events reported in the news. It provides two alternatives, one supervised, the other unsupervised to track events in the texts.

This software enables to damage texts written in any natural language by applying OCR degradation (phantom characters, character degradation, etc.) and by over-segmenting texts (this means splitting regularly the texts in equal parts).

This is useful to reproduce common errors found in historical documents when historical data is missing.

This software is used to compute TF IDF weighting from texts that are based on the document_tracking_resources format. Vectors and weightings are computed thanks to a resource file that contains a representation of the language used in the same context as the text to weight (news features to weight texts published in the news).

This software is used to compute dense vectorisations (sentence embeddings) of sequences of sentences of natural text. It is able to handle multilingual documents until the model used is a multilingual one. This relies on the S-BERT architecture, software and models (https://www.sbert.net/). It computes dense vector representations for tokens, lemmas, entities, etc. of your datasets.

Process documents in order to extract tokens, lemmas and named entities from texts. This software depends on spaCy (https://spacy.io/) in order to extract text features and recognise the inner elements.

Annotation tool to check whether the annotation of a Corpus (document_tracking_resources) are correct and truthful.

Projet basculé sur gitlab-dsi (https://gitlab-dsi.univ-lr.fr/dsi-soft/audiovisuel/linx)