Noteebook_Capitalisation_RAG_Platform_20250220.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "cdd5bdb3-6199-4289-9d1a-cb3c3839bb5a",
   "metadata": {},
   "source": [
    "# Controversy RaG : classification automatique de publications scientifiques"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d9747cc5-0103-4245-9567-af7eb0224a83",
   "metadata": {},
   "source": [
    "Ce notebook reprend les enseignements des expérimentations menées dans le cadre d'un projet d'automatisation de l'identification de publications scientifiques appartenant au domaine des Sciences et Recherches Participatives. \n",
    "\n",
    "Il capitalise plusieurs mois d'expérimentations en matière d'apprentissage machine (few-shot learning) reposant sur l'utilisation des modèles de langues de dimensions variées."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6e862775-7017-4e9e-a865-020c8c048ccf",
   "metadata": {},
   "source": [
    "## Description du projet"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78632b41-fae1-4998-89c2-11f284b54b1b",
   "metadata": {},
   "source": [
    "## Problématique\n",
    "\n",
    "Lorsque des experts doivent décider si un document appartient ou pas au domaine des Sciences et Recherches Participatives, ils se basent sur une série d'indices textuels qui malheureusement, en dehors de l'identification de termes clés, ne sont pas explicités. Cela s'explique en partie par le fait que les alignements cognitifs qui se font entre ce qu'ils lisent et percoivent d'un ensemble de résumés et de mots-clés auteurs et la représentation mentale qu'ils se font du champ à délinéer, sont largement inconscients.\n",
    "\n",
    "Cela signifie que lorsque l'on veut traduire ces mécanismes de catégorisation en algorithme d'apprentissage on ne dispose en réalité que de peu de matériaux. \n",
    "\n",
    "Dans notre cas, ces matériaux se résument à une série de mot-clés. Or, la présence ou l'absence de mots-clés ne suffisent pas à eux seuls à catégoriser l'appartenance d'un texte à un domaine. \n",
    "\n",
    "[donner exemple]\n",
    "\n",
    "L'objectif de ces expérimentations était de voir si en utilisant des modèles de langue, un ensemble minimal de données annotées, on est en mesure de recoder, plus ou moins parfaitement "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6fe8594e-8b93-4d80-af3c-fb8f5bdee515",
   "metadata": {},
   "source": [
    "## Code - Experimentations"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3730f089-7102-4783-968a-d2b4301d4fe6",
   "metadata": {},
   "source": [
    "### Installation des paramètres d'environnement"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "9d8cf842-3f72-4300-8988-3bac5e814fad",
   "metadata": {},
   "outputs": [],
   "source": [
    "### Chargement des librairies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "9f63996b-32c3-498e-babb-9d7e7d97f8c4",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### A. Librairies python pour la gestion de l'environnement de travail\n",
    "import os\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "2d8b00bb-0129-4c9f-bbea-be259942f6c7",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### B. Librairies python pour la manipulation de string\n",
    "import ast\n",
    "import json\n",
    "import re"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "9d7c2d9b-e095-41cc-8910-d087a66ebb01",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO: Pandarallel will run on 6 workers.\n",
      "INFO: Pandarallel will use Memory file system to transfer data between the main process and workers.\n"
     ]
    }
   ],
   "source": [
    "#### C. Librairies python pour la manipulation de dataframes\n",
    "import numpy as np\n",
    "from pandarallel import pandarallel\n",
    "pandarallel.initialize(nb_workers=6,progress_bar=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "59e02aee-d698-4748-b2ca-4655a2421705",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### D. Librairies python pour la production de datavisualisation et \n",
    "#### la manipulation de données vectorielles\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn.cluster import KMeans\n",
    "from sklearn.metrics import silhouette_score\n",
    "from sklearn.metrics.pairwise import cosine_distances\n",
    "from sklearn.neighbors import NearestNeighbors\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "ab0479ab-3e8e-4c5e-bb46-c5f062112a39",
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "1e85a083-b5e8-4e15-a422-d934bc06c2d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### E. Librairies python dédiées au NLP\n",
    "import rank_bm25\n",
    "from rank_bm25 import BM25Okapi\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sentence_transformers import SentenceTransformer\n",
    "from transformers import BertTokenizer,BertTokenizerFast, BertModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "6570ccbf-1822-4c95-ba42-c22d0d4287b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### F. Librairies python pour la manipulation de modèles de langue \n",
    "#### et la création de pipelines de traitements\n",
    "from langchain_experimental.text_splitter import SemanticChunker\n",
    "from langchain_huggingface import HuggingFaceEmbeddings\n",
    "from langchain.retrievers import BM25Retriever, EnsembleRetriever\n",
    "from langchain_core.documents import Document\n",
    "from langchain_chroma import Chroma\n",
    "from langchain_huggingface import HuggingFaceEmbeddings\n",
    "from chromadb.utils import embedding_functions\n",
    "import torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "ea3ccc29-df60-4a1d-8796-a8dc0a894fcc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda:0\n"
     ]
    }
   ],
   "source": [
    "### On vérifie que l'on travaille bien avec le gpu\n",
    "device = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\n",
    "print(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3a65f885-b6c8-4083-a05b-a1ac81fa4b78",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### G. Modules maison\n",
    "##### tok_func est un module (script python ici) qui rassemble dans un même fichier\n",
    "##### l'ensembles des fonctions de pre-processing des données textuelles (nettoyage, \n",
    "##### standardisation, tokenization, lemmatisation, ...)\n",
    "from tok_func import *\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "f435e71e-3211-4be0-a67b-08d841282f40",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### Pré-chargment de modèles (ne pas toucher)\n",
    "##### Pour éviter certains warnings dû à l'utilisation d'opérations\n",
    "##### d'encodage de données textuelles se faisant dans des fonctions\n",
    "##### on précharge avant l'utilisation de ces fonctions (dans le script tok_func)\n",
    "##### les modèles de langue à utiliser.\n",
    "##### Attention le modèle Qwen Instruct est très lourd. Sur votre machine vous devrez \n",
    "##### sûrement le modifier pour un modèle plus léger.\n",
    "###https://huggingface.co/sentence-transformers/allenai-specter\n",
    "docmodel= None\n",
    "kwrdmdl = None\n",
    "spectermodel = SentenceTransformer(\n",
    "   \"sentence-transformers/allenai-specter\", trust_remote_code=True\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7d76c76c-282a-431a-8f61-4c5eb1f74074",
   "metadata": {},
   "source": [
    "### Fonctions utilisées "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "00ac6224-198f-478e-816d-4cb54d941d99",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_sbert_embedding(document, model_name='all-MiniLM-L6-v2'):\n",
    "    '''Function takes as input raw text and return its bert embeddings.\n",
    "    It uses sentence Bert as we focus at the text level and not the token level'''\n",
    "    global docmodel\n",
    "    # Load the SBERT model only if it hasn't been loaded yet\n",
    "    if docmodel is None:\n",
    "        docmodel = SentenceTransformer(model_name)\n",
    "\n",
    "    # Get the embedding\n",
    "    embedding = docmodel.encode(document, convert_to_numpy=True)\n",
    "    \n",
    "    return embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "b18eb638-7ef6-4618-bb86-1da8e0c14ffb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO: Pandarallel will run on 30 workers.\n",
      "INFO: Pandarallel will use Memory file system to transfer data between the main process and workers.\n"
     ]
    }
   ],
   "source": [
    "from pandarallel import pandarallel\n",
    "pandarallel.initialize(nb_workers=30,progress_bar=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6feede94-3789-4875-bfed-3b6eb9fc3b60",
   "metadata": {},
   "source": [
    "### Import et PreProcessing des données"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "c2cec83e-3a54-4303-ac83-d77738da57d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### le dataframe keywords_df contient la liste des mots-clefs repérés par les experts \n",
    "#### pour la construction de la requête d'interrogation des platformes bibliographiques\n",
    "#### ainsi que des mots-clefs supplémentaires repérés au cours de leur examen des ensembles\n",
    "#### titres+résumés\n",
    "\n",
    "keywords_df = pd.read_csv(\"/home/trix/Dev/classification/Data/corpuskeywordsV2.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "b3b369fa-c42d-4918-9d57-ee3445a71544",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1156\n",
      "1156\n"
     ]
    }
   ],
   "source": [
    "#### le dataframe expert_df constitue le corpus d'apprentissage et d'entraînement à partir\n",
    "#### duquel les expérimentations de few-shot learning seront menées.\n",
    "#### Ce dataset a été annoté par des experts, et augmentés de données afin d'accroître le\n",
    "#### nombre de publications hors-champ.\n",
    "\n",
    "expert_df = pd.read_excel(\"/home/trix/Dev/classification/Data/Annotated_Complemented_ExpertDataset_20241106.xlsx\")\n",
    "print(len(expert_df))\n",
    "\n",
    "#### vérifions qu'il n'y ait pas de publications en double dans ce dataset \n",
    "expert_df.drop_duplicates(subset='cle_UT')\n",
    "print(len(expert_df))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "b9314924-9b61-4193-85a8-4e05565c0a87",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Composition du dataset pour l'entraînement et les tests des expérimentations de classification:\n",
      "\n",
      "\tNombre de publications du dataset:\n",
      "\t1156\n",
      "\n",
      "\tNombre de publications dans le champ:\n",
      "\t242\n",
      "\n",
      "\tNombre de publications hors-champ:\n",
      "\t883\n",
      "\n",
      "\tNombre de publications ambigües:\n",
      "\t30\n"
     ]
    }
   ],
   "source": [
    "print(\"Composition du dataset pour l'entraînement et les tests des expérimentations de classification:\")\n",
    "print(f\"\\n\\tNombre de publications du dataset:\\n\\t{len(expert_df)}\")\n",
    "print(f\"\\n\\tNombre de publications dans le champ:\\n\\t{len(expert_df[expert_df.label == 'yes'])}\")\n",
    "print(f\"\\n\\tNombre de publications hors-champ:\\n\\t{len(expert_df[expert_df.label == 'no'])}\")\n",
    "print(f\"\\n\\tNombre de publications ambigües:\\n\\t{len(expert_df[expert_df.label == 'ambiguous'])}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "fb0ab3a4-4702-4fde-84d0-fd01df5ce3e4",
   "metadata": {},
   "outputs": [],
   "source": [
    "#### Dans notre cas il existe en réalité une troisime catégorie identifiée \n",
    "#### par les experts: ambiguës. Cetta catégorie est toutefois trop peu importante\n",
    "#### pour que l'on puisse s'en servir (résultat de précédentes expérimentations).\n",
    "#### On considère donc qu'elles sont hors-champ. Pour ne pas perdre leur trace\n",
    "#### on effectue une copie du dataframe original.\n",
    "\n",
    "expert_copy = expert_df.copy()\n",
    "expert_copy['label'] = expert_copy.label.apply(lambda x: 'participative science' if x == 'yes' else x)\n",
    "expert_copy['label'] = expert_copy.label.apply(lambda x: 'non-participative science' if x == 'no' else x)\n",
    "expert_copy['label'] = expert_copy.label.apply(lambda x: 'non-participative science' if x == 'ambiguous' else x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "092d0945-e874-41b3-8397-71d015ecff1b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Composition du dataset pour l'entraînement et le test du classificateur:\n",
      "\n",
      "\tNombre de publications du dataset:\n",
      "\t1156\n",
      "\n",
      "\tNombre de publications dans le champ:\n",
      "\t242\n",
      "\n",
      "\tNombre de publications hors-champ:\n",
      "\t913\n",
      "\n",
      "\tNombre de publications ambigües:\n",
      "\t0\n"
     ]
    }
   ],
   "source": [
    "print(\"Composition du dataset pour l'entraînement et le test du classificateur:\")\n",
    "print(f\"\\n\\tNombre de publications du dataset:\\n\\t{len(expert_copy)}\")\n",
    "print(f\"\\n\\tNombre de publications dans le champ:\\n\\t{len(expert_copy[expert_copy.label == 'participative science'])}\")\n",
    "print(f\"\\n\\tNombre de publications hors-champ:\\n\\t{len(expert_copy[expert_copy.label == 'non-participative science'])}\")\n",
    "print(f\"\\n\\tNombre de publications ambigües:\\n\\t{len(expert_copy[expert_copy.label == 'ambiguous'])}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9aab6dfa-e7ee-4df9-8566-bb51fddc1381",
   "metadata": {},
   "source": [
    "Comme on peut le voir, dans le corpus qui va nous servir de dataset d'entraînement et de test, 79% des publications annotées sont hors-champ; 21% dans le champ."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "a7fcc0cc-708c-4973-941c-87010bff5395",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9ee82e88c58540a281549d038a23cbf2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HBox(children=(IntProgress(value=0, description='0.00%', max=39), Label(value='0 / 39'))), HBox…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e6b92ba89fb84a31a58fb17fdb36ac3e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HBox(children=(IntProgress(value=0, description='0.00%', max=39), Label(value='0 / 39'))), HBox…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "75138e95632a4c6db780357d72fcf0cd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HBox(children=(IntProgress(value=0, description='0.00%', max=39), Label(value='0 / 39'))), HBox…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "ename": "KeyError",
     "evalue": "\"['Unnamed: 0'] not found in axis\"",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "File \u001b[0;32m<timed exec>:12\u001b[0m\n",
      "File \u001b[0;32m~/pythonenvs/envs/llm_rag/lib/python3.11/site-packages/pandas/core/frame.py:5581\u001b[0m, in \u001b[0;36mDataFrame.drop\u001b[0;34m(self, labels, axis, index, columns, level, inplace, errors)\u001b[0m\n\u001b[1;32m   5433\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21mdrop\u001b[39m(\n\u001b[1;32m   5434\u001b[0m     \u001b[38;5;28mself\u001b[39m,\n\u001b[1;32m   5435\u001b[0m     labels: IndexLabel \u001b[38;5;241m|\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m   5442\u001b[0m     errors: IgnoreRaise \u001b[38;5;241m=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mraise\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m   5443\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m DataFrame \u001b[38;5;241m|\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m   5444\u001b[0m \u001b[38;5;250m    \u001b[39m\u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m   5445\u001b[0m \u001b[38;5;124;03m    Drop specified labels from rows or columns.\u001b[39;00m\n\u001b[1;32m   5446\u001b[0m \n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m   5579\u001b[0m \u001b[38;5;124;03m            weight  1.0     0.8\u001b[39;00m\n\u001b[1;32m   5580\u001b[0m \u001b[38;5;124;03m    \"\"\"\u001b[39;00m\n\u001b[0;32m-> 5581\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdrop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   5582\u001b[0m \u001b[43m        \u001b[49m\u001b[43mlabels\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mlabels\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5583\u001b[0m \u001b[43m        \u001b[49m\u001b[43maxis\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43maxis\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5584\u001b[0m \u001b[43m        \u001b[49m\u001b[43mindex\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mindex\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5585\u001b[0m \u001b[43m        \u001b[49m\u001b[43mcolumns\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mcolumns\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5586\u001b[0m \u001b[43m        \u001b[49m\u001b[43mlevel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mlevel\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5587\u001b[0m \u001b[43m        \u001b[49m\u001b[43minplace\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minplace\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5588\u001b[0m \u001b[43m        \u001b[49m\u001b[43merrors\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43merrors\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   5589\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/pythonenvs/envs/llm_rag/lib/python3.11/site-packages/pandas/core/generic.py:4788\u001b[0m, in \u001b[0;36mNDFrame.drop\u001b[0;34m(self, labels, axis, index, columns, level, inplace, errors)\u001b[0m\n\u001b[1;32m   4786\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m axis, labels \u001b[38;5;129;01min\u001b[39;00m axes\u001b[38;5;241m.\u001b[39mitems():\n\u001b[1;32m   4787\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m labels \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[0;32m-> 4788\u001b[0m         obj \u001b[38;5;241m=\u001b[39m \u001b[43mobj\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_drop_axis\u001b[49m\u001b[43m(\u001b[49m\u001b[43mlabels\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43maxis\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mlevel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mlevel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43merrors\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43merrors\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   4790\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m inplace:\n\u001b[1;32m   4791\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_update_inplace(obj)\n",
      "File \u001b[0;32m~/pythonenvs/envs/llm_rag/lib/python3.11/site-packages/pandas/core/generic.py:4830\u001b[0m, in \u001b[0;36mNDFrame._drop_axis\u001b[0;34m(self, labels, axis, level, errors, only_slice)\u001b[0m\n\u001b[1;32m   4828\u001b[0m         new_axis \u001b[38;5;241m=\u001b[39m axis\u001b[38;5;241m.\u001b[39mdrop(labels, level\u001b[38;5;241m=\u001b[39mlevel, errors\u001b[38;5;241m=\u001b[39merrors)\n\u001b[1;32m   4829\u001b[0m     \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 4830\u001b[0m         new_axis \u001b[38;5;241m=\u001b[39m \u001b[43maxis\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdrop\u001b[49m\u001b[43m(\u001b[49m\u001b[43mlabels\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43merrors\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43merrors\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   4831\u001b[0m     indexer \u001b[38;5;241m=\u001b[39m axis\u001b[38;5;241m.\u001b[39mget_indexer(new_axis)\n\u001b[1;32m   4833\u001b[0m \u001b[38;5;66;03m# Case for non-unique axis\u001b[39;00m\n\u001b[1;32m   4834\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n",
      "File \u001b[0;32m~/pythonenvs/envs/llm_rag/lib/python3.11/site-packages/pandas/core/indexes/base.py:7070\u001b[0m, in \u001b[0;36mIndex.drop\u001b[0;34m(self, labels, errors)\u001b[0m\n\u001b[1;32m   7068\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m mask\u001b[38;5;241m.\u001b[39many():\n\u001b[1;32m   7069\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m errors \u001b[38;5;241m!=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mignore\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n\u001b[0;32m-> 7070\u001b[0m         \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mKeyError\u001b[39;00m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mlabels[mask]\u001b[38;5;241m.\u001b[39mtolist()\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m not found in axis\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m   7071\u001b[0m     indexer \u001b[38;5;241m=\u001b[39m indexer[\u001b[38;5;241m~\u001b[39mmask]\n\u001b[1;32m   7072\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdelete(indexer)\n",
      "\u001b[0;31mKeyError\u001b[0m: \"['Unnamed: 0'] not found in axis\""
     ]
    }
   ],
   "source": [
    "%%time\n",
    "#### Plusieurs opérations sont réalisées ici:\n",
    "#### - découpage des données textuelles (tritres+abstracts) en phrases (raw sentences),\n",
    "#### - nettoyage de ces phrases (cleaned_sentences),\n",
    "#### - transformation des keywords auteur en liste de mots-clés (changement de type de données),\n",
    "\n",
    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"True\"\n",
    "expert_copy['raw_sentences'] = expert_copy.tires.apply(lambda x: get_sentences(x,tokenized=False,rawsent = True))\n",
    "expert_copy['cleaned_sentences'] = expert_copy.tires.parallel_apply(lambda x: get_sentences(x,tokenized=False, rawsent = False))\n",
    "expert_copy['tokenized_doc'] = expert_copy.tires.parallel_apply(lambda x: get_sentences(x, tokenized=True))\n",
    "expert_copy['keywords'] = expert_copy.keywords.apply(lambda x: ast.literal_eval(x) if x!= \"Null\" else [\"Null\"])\n",
    "expert_copy['participatory_keywords'] = expert_copy.cleaned_sentences.parallel_apply(lambda x: extract_targets(x))\n",
    "expert_copy = expert_copy.drop(columns=['Unnamed: 0'],axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "d290775f-760a-4b32-b96d-fdfe9d982dd3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Index</th>\n",
       "      <th>cle_UT</th>\n",
       "      <th>doi</th>\n",
       "      <th>tires</th>\n",
       "      <th>keywords</th>\n",
       "      <th>year</th>\n",
       "      <th>label</th>\n",
       "      <th>references</th>\n",
       "      <th>raw_sentences</th>\n",
       "      <th>cleaned_sentences</th>\n",
       "      <th>tokenized_doc</th>\n",
       "      <th>participatory_keywords</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>11</td>\n",
       "      <td>WOS:000928116000001</td>\n",
       "      <td>10.1177/16094069221137492</td>\n",
       "      <td>Tailoring Cognitive Mapping Analysis Methods t...</td>\n",
       "      <td>[participatory_research, social_learning, coll...</td>\n",
       "      <td>2022</td>\n",
       "      <td>participative science</td>\n",
       "      <td>001-076-538-431-45X,001-162-706-708-552,004-42...</td>\n",
       "      <td>tailor cognitive mapping analysis methods diff...</td>\n",
       "      <td>tailor cognitive mapping analysis methods diff...</td>\n",
       "      <td>[tailor, cognitive, mapping, analysis, methods...</td>\n",
       "      <td>participatory_process</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>12</td>\n",
       "      <td>WOS:000895513400001</td>\n",
       "      <td>10.3390/land11112095</td>\n",
       "      <td>Who and Where Are the Observers behind Biodive...</td>\n",
       "      <td>[citizen_science, spatial_bias, observer_profi...</td>\n",
       "      <td>2022</td>\n",
       "      <td>participative science</td>\n",
       "      <td>000-826-091-294-906,002-078-837-621-902,003-66...</td>\n",
       "      <td>observer biodiversity citizen_science data. ef...</td>\n",
       "      <td>observer biodiversity citizen_science data. ef...</td>\n",
       "      <td>[observer, biodiversity, citizen_science, data...</td>\n",
       "      <td>behavior_volunteer_engagement;citizen_science;...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>14</td>\n",
       "      <td>WOS:000880001000004</td>\n",
       "      <td>10.1016/j.ejsobi.2022.103449</td>\n",
       "      <td>Gut content metabarcoding and citizen science ...</td>\n",
       "      <td>[citizen_science, earthworms, flatworms, gut_c...</td>\n",
       "      <td>2022</td>\n",
       "      <td>participative science</td>\n",
       "      <td>000-069-184-790-014,001-257-411-314-46X,003-53...</td>\n",
       "      <td>gut content metabarcoding citizen_science reve...</td>\n",
       "      <td>gut content metabarcoding citizen_science reve...</td>\n",
       "      <td>[gut, content, metabarcoding, citizen_science,...</td>\n",
       "      <td>citizen_science;volunteer_reporting</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  Index               cle_UT                           doi  \\\n",
       "0    11  WOS:000928116000001     10.1177/16094069221137492   \n",
       "1    12  WOS:000895513400001          10.3390/land11112095   \n",
       "2    14  WOS:000880001000004  10.1016/j.ejsobi.2022.103449   \n",
       "\n",
       "                                               tires  \\\n",
       "0  Tailoring Cognitive Mapping Analysis Methods t...   \n",
       "1  Who and Where Are the Observers behind Biodive...   \n",
       "2  Gut content metabarcoding and citizen science ...   \n",
       "\n",
       "                                            keywords  year  \\\n",
       "0  [participatory_research, social_learning, coll...  2022   \n",
       "1  [citizen_science, spatial_bias, observer_profi...  2022   \n",
       "2  [citizen_science, earthworms, flatworms, gut_c...  2022   \n",
       "\n",
       "                   label                                         references  \\\n",
       "0  participative science  001-076-538-431-45X,001-162-706-708-552,004-42...   \n",
       "1  participative science  000-826-091-294-906,002-078-837-621-902,003-66...   \n",
       "2  participative science  000-069-184-790-014,001-257-411-314-46X,003-53...   \n",
       "\n",
       "                                       raw_sentences  \\\n",
       "0  tailor cognitive mapping analysis methods diff...   \n",
       "1  observer biodiversity citizen_science data. ef...   \n",
       "2  gut content metabarcoding citizen_science reve...   \n",
       "\n",
       "                                   cleaned_sentences  \\\n",
       "0  tailor cognitive mapping analysis methods diff...   \n",
       "1  observer biodiversity citizen_science data. ef...   \n",
       "2  gut content metabarcoding citizen_science reve...   \n",
       "\n",
       "                                       tokenized_doc  \\\n",
       "0  [tailor, cognitive, mapping, analysis, methods...   \n",
       "1  [observer, biodiversity, citizen_science, data...   \n",
       "2  [gut, content, metabarcoding, citizen_science,...   \n",
       "\n",
       "                              participatory_keywords  \n",
       "0                              participatory_process  \n",
       "1  behavior_volunteer_engagement;citizen_science;...  \n",
       "2                citizen_science;volunteer_reporting  "
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "expert_copy.head(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b9e1da04-1534-4050-890b-b192437bf2ff",
   "metadata": {},
   "source": [
    "### Premiers éléments d'analyses"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff9762f9-924e-4e17-ae4e-fcc8ae51b4f9",
   "metadata": {},
   "source": [
    "Dans un premier temps, on souhaite regarder si parmi la liste des mots-clés des experts certains sont plus représentatifs des publications appartenant au champ et si d'autres, au contraire sont plutôt ambiguës. \n",
    "\n",
    "Pour cela on va regarder la fréquence de distribution de chacun des mots-clés expert au sein des titres+abstracts et regarder comment les résultats obtenus permettent ou pas de discriminer les publications dans ou hors-champ. Si nous ne ferons pas un usage immédiat de ces résultats cela nous donne cependant un indice de saillance de ces mots-clés, et nous permettra plus tard de réduire le bruit qui met en défaut les modèles de langues au moment de la classification."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "cf6eb010-959c-4971-9e95-a1d7109e4fa2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 6.24 s, sys: 4.81 ms, total: 6.25 s\n",
      "Wall time: 6.24 s\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Expert Keyword</th>\n",
       "      <th>Mean Freq_Part</th>\n",
       "      <th>Mean Freq_NonPart</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>citizen_science</td>\n",
       "      <td>0.342975</td>\n",
       "      <td>0.002191</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>89</th>\n",
       "      <td>participatory_approach</td>\n",
       "      <td>0.276860</td>\n",
       "      <td>0.013143</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>98</th>\n",
       "      <td>participatory_process</td>\n",
       "      <td>0.119835</td>\n",
       "      <td>0.007667</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>101</th>\n",
       "      <td>participatory_research</td>\n",
       "      <td>0.115702</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>44</th>\n",
       "      <td>codesign_process</td>\n",
       "      <td>0.103306</td>\n",
       "      <td>0.058050</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>co_design_process</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>79</th>\n",
       "      <td>onfarm_participatory</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>multiactor_coinnovation_workshop</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76</th>\n",
       "      <td>multiactor_co-innovation_workshop</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>68</th>\n",
       "      <td>innovation_codesigne</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>136 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                        Expert Keyword  Mean Freq_Part  Mean Freq_NonPart\n",
       "18                     citizen_science        0.342975           0.002191\n",
       "89              participatory_approach        0.276860           0.013143\n",
       "98               participatory_process        0.119835           0.007667\n",
       "101             participatory_research        0.115702           0.000000\n",
       "44                    codesign_process        0.103306           0.058050\n",
       "..                                 ...             ...                ...\n",
       "28                   co_design_process        0.000000           0.000000\n",
       "79                onfarm_participatory        0.000000           0.000000\n",
       "77    multiactor_coinnovation_workshop        0.000000           0.000000\n",
       "76   multiactor_co-innovation_workshop        0.000000           0.000000\n",
       "68                innovation_codesigne        0.000000           0.000000\n",
       "\n",
       "[136 rows x 3 columns]"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "kscores=[]\n",
    "for k in keywords_df.Keywords.to_list():\n",
    "    partokenlist = expert_copy['tokenized_doc'][expert_copy.label == 'participative science'].to_list()\n",
    "    partbm25 = BM25Okapi(partokenlist)\n",
    "    nonpartokenlist = expert_copy['tokenized_doc'][expert_copy.label == 'non-participative science'].to_list()\n",
    "    nonpartbm25 = BM25Okapi(nonpartokenlist)\n",
    "    \n",
    "    kpartcount = [tklist.count(k) for tklist in partokenlist]\n",
    "    partbm25_scores = partbm25.get_scores(k)\n",
    "    kpartmean = sum(kpartcount)/len(partokenlist)\n",
    "    partbm25mean = sum(partbm25_scores)/len(partbm25_scores)\n",
    "    \n",
    "    knonpartcount = [tklist.count(k) for tklist in nonpartokenlist]\n",
    "    nonpartbm25_scores = nonpartbm25.get_scores(k)\n",
    "    knonpartmean = sum(knonpartcount)/len(nonpartokenlist)\n",
    "    nonpartbm25mean = sum(nonpartbm25_scores)/len(nonpartbm25_scores)\n",
    "    kscores.append([k,kpartmean,knonpartmean])\n",
    "\n",
    "kwrd_freq_df = pd.DataFrame(data=kscores,columns=['Expert Keyword','Mean Freq_Part','Mean Freq_NonPart'])\n",
    "kwrd_freq_df.sort_values(by='Mean Freq_Part',ascending = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "080de2e0-7db5-4a1e-bad6-9d8767f82ff1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "kwrd_freq_df.boxplot(column=['Mean Freq_Part', 'Mean Freq_NonPart'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e36ce5ea-c0e8-4df4-9034-6d21a9642a5d",
   "metadata": {},
   "source": [
    "On s'apperçoit que parmi la liste des mots-clés experts retournés, un grand nombre d'entre eux n'apparaissent pas dans les titres-mots clés concaténés. En dehors de ceux pour lesquels la fréquence d'apparition est positive, ils semblent donc peu discriminant pour les publications de ce champ. Ce résultat apparaît d'autant plus contre-intuitif que ce sont ces mêmes mots clefs qui ont été utilisés pour la réalisation de la requête.\n",
    "\n",
    "Il conviendrait ici d'extraire les mots clefs représentatifs de ces publications à partir d'un algorithme comme BM25. [à implémenter avec service alvis](https://text-mining-dev.migale.inrae.fr/demo/kes/)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "02d30200-ed8e-4a74-8c5c-ddb7b9ec3cb6",
   "metadata": {},
   "source": [
    "On peut se demander si à l'inverse les mots-clés auteurs sont plus discriminants. On réitère pour ce faire l'opération précédente mais en utilisant les mots clés des auteurs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "4f83e559-db5a-4cf9-9eac-77195b6e0fbf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5919\n",
      "4541\n",
      "CPU times: user 2.64 s, sys: 4.88 ms, total: 2.64 s\n",
      "Wall time: 2.64 s\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Author Keyword</th>\n",
       "      <th>Mean Freq_Part</th>\n",
       "      <th>Mean Freq_NonPart</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>672</th>\n",
       "      <td>citizen_science</td>\n",
       "      <td>0.024011</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3047</th>\n",
       "      <td>participatory_research</td>\n",
       "      <td>0.016949</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>126</th>\n",
       "      <td>agroecology</td>\n",
       "      <td>0.014124</td>\n",
       "      <td>0.004225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3029</th>\n",
       "      <td>participatory_approach</td>\n",
       "      <td>0.012006</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4065</th>\n",
       "      <td>sustainability</td>\n",
       "      <td>0.009887</td>\n",
       "      <td>0.000667</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1737</th>\n",
       "      <td>gestational_diabetes</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1738</th>\n",
       "      <td>gestational_weight_gain</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1739</th>\n",
       "      <td>gh70_family</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1742</th>\n",
       "      <td>gibel_carp</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4540</th>\n",
       "      <td>zymoseptoria_tritici</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4541 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "               Author Keyword  Mean Freq_Part  Mean Freq_NonPart\n",
       "672           citizen_science        0.024011           0.000222\n",
       "3047   participatory_research        0.016949           0.000222\n",
       "126               agroecology        0.014124           0.004225\n",
       "3029   participatory_approach        0.012006           0.000000\n",
       "4065           sustainability        0.009887           0.000667\n",