HaRe.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "HaRe\n",
    "====\n",
    "\n",
    "HaRe (Harassment Recognizer) is a command line tool and Python library to automatically detect harassment as it happens (real-time) with the help of machine learning techniques. In this notebook we will look at the Python library.\n",
    "\n",
    "Basic usage: monitoring conversations\n",
    "--------------------------------------------------------\n",
    "\n",
    "The easiest way to use HaRe is by simply loading a pretrained HaRe model. You can load it by simply doing:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from hare import load_pretrained\n",
    "moba_hare = load_pretrained('hare/pretrained/moba')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can then use this object to monitor conversations in progress. Let's start a conversation and ask HaRe to monitor it:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from hare import Conversation\n",
    "\n",
    "convo = Conversation()\n",
    "moba_hare.add_conversation(convo)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can add utterances to this conversation as they become available, and at any point in time, you can then request the current status of the conversation according to this HaRe model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'a': 0.008861654438078403, 'b': 0.005346317775547504}"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "convo.add_utterance(speaker='a',content='hello')\n",
    "convo.add_utterance(speaker='b',content='hi everyone')\n",
    "moba_hare.get_status()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can also add multiple sentences at once (for example a whole conversation if it has already finished), and see what HaRe's best guess was at every moment in time."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "a: hello\n",
      "{'a': 0.008861654438078403}\n",
      "---\n",
      "\n",
      "b: hi everyone\n",
      "{'a': 0.008861654438078403, 'b': 0.005346317775547504}\n",
      "---\n",
      "\n",
      "a: good luck\n",
      "{'a': 0.010540383867919445, 'b': 0.005346317775547504}\n",
      "---\n",
      "\n",
      "c: noob team\n",
      "{'a': 0.010540383867919445, 'b': 0.005346317775547504, 'c': 0.07421243190765381}\n",
      "---\n",
      "\n",
      "c: u guys suck\n",
      "{'a': 0.010540383867919445, 'b': 0.005346317775547504, 'c': 0.5076202750205994}\n",
      "---\n",
      "\n",
      "b: ur a loser yourself\n",
      "{'a': 0.010540383867919445, 'b': 0.010474762879312038, 'c': 0.5076202750205994}\n",
      "---\n",
      "\n",
      "a: take it easy guys\n",
      "{'a': 0.012263094075024128, 'b': 0.010474762879312038, 'c': 0.5076202750205994}\n",
      "---\n",
      "\n",
      "c: noob\n",
      "{'a': 0.012263094075024128, 'b': 0.010474762879312038, 'c': 0.5150563716888428}\n",
      "---\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from hare import Utterance\n",
    "\n",
    "convo.add_utterances([Utterance(speaker='a',content='good luck'),\n",
    "                      Utterance(speaker='c',content='noob team'),\n",
    "                      Utterance(speaker='c',content='u guys suck'),\n",
    "                      Utterance(speaker='b',content='ur a loser yourself'),\n",
    "                      Utterance(speaker='a',content='take it easy guys'),\n",
    "                      Utterance(speaker='c',content='noob')])\n",
    "\n",
    "moba_hare.visualize_history_for_conversation()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Or in a graph:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "from hare.visualize import visualize_toxicity_for_one_conversation\n",
    "    \n",
    "visualize_toxicity_for_one_conversation(moba_hare)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If you add multiple conversations for Hare to monitor, you will need to specify the conversation index when asking for the status:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'a': 0.017604785040020943, 'b': 0.005346317775547504}"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "second_convo = Conversation()\n",
    "\n",
    "second_convo.add_utterance(speaker='a',content='hello')\n",
    "second_convo.add_utterance(speaker='b',content='hi everyone')\n",
    "second_convo.add_utterance(speaker='a',content='go away loser!')\n",
    "\n",
    "moba_hare.add_conversation(second_convo)\n",
    "moba_hare.get_status(id=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Evaluating \n",
    "----------\n",
    "If you have a labeled dataset (that is: for each conversation an indication which participants are considered toxic), HaRe can calculate to what extent its judgments match the labels. The labels can either be the default 0 (not toxic) or 1 (maximally toxic). Labeling speaker c in the first conversation above would look like this:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "convo.label_speaker('c',1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There are several evaluation metrics, depending on what is important to you (detecting ALL harassment, detecting harassment quickly, no false positives, etc). You can ask a HaRe object to output these aggregate scores directly, but we can also visualize them similar to how we visualized the estimates for a single conversation above. You can use these metrics to find out which HaRe object suits your goals best. \n",
    "\n",
    "Let's do an experiment. Besides the default machine-learning based engine, called a 'brain' in the HaRe framework, it's also possible to simply detect toxicity based on a wordlist using a so-called `DictBasedBrain`. Below we take a pretrained machine-learning based classifier (named `moba_hare`), and compare it with two Hare objects using such a `DictBasedBrain` (named `insult_hare` and `swear_hare`)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "from hare import Hare\n",
    "from hare.brain import DictBasedBrain\n",
    "\n",
    "moba_hare = load_pretrained('hare/pretrained/moba')\n",
    "moba_hare.name = 'machine-learning'\n",
    "\n",
    "insult_hare = Hare(name='insults')\n",
    "insult_hare.brain = DictBasedBrain()\n",
    "insult_hare.brain.bad_words = ['noob','newb','n00b','fag','loser']\n",
    "\n",
    "swear_hare = Hare(name='swear_words')\n",
    "swear_hare.brain = DictBasedBrain()\n",
    "swear_hare.brain.bad_words = ['fuck','fck','fuk','shit','stfu','wtf','suck']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "HaRe comes with 8 fictional example conversations that are typical for a videogame environment. Let's give these to our three Hare objects to monitor."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded 8 conversations\n"
     ]
    }
   ],
   "source": [
    "from hare import load_example_conversations\n",
    "\n",
    "example_conversations = load_example_conversations()\n",
    "print('Loaded',len(example_conversations),'conversations')\n",
    "\n",
    "for conversation in example_conversations:\n",
    "    insult_hare.add_conversation(conversation)\n",
    "    swear_hare.add_conversation(conversation)\n",
    "    moba_hare.add_conversation(conversation)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we have multiple labeled conversations and HaRe objects to analyze them, let's visualize the evaluation measure 'Area under the (ROC) Curve' to see how they perform. An AUC of 0.5 means that the HaRe object is performing at chance level, an AUC of 1 means that it's perfect."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "from hare.visualize import visualize_accuracy_during_conversations, visualize_auc_during_conversations, visualize_fscore_during_conversations\n",
    "\n",
    "visualize_auc_during_conversations([insult_hare,swear_hare,moba_hare])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We see that the estimates are poor in the beginning of conversations, when there is almost no information, and that this gradually gets better as more people speak. The machine learning generally identifies the toxic player earlier in the conversation, but the word list model, in particular the one with swear words, picks up later. Also note that scores can decrease again over time, as a system might get confused with utterances at a later point in the conversation.\n",
    "\n",
    "In the graph above, a player is 'identified' as toxic if it has an toxicity estimate above 0.5 . Of course, you can play with this: if you set this threshold to a lower value, there will be more positives (true positives, but also false positives), and if you set it to a higher value, there will be more negatives (true negatives, but also false negatives). \n",
    "\n",
    "Instead of visualizing a score over time, we can also visualize the score with various of these thresholds. The  the ROC curve and the precision/recall graph are two common ways of doing this. In the first graph below, lines closer to the upper left corner are better, and in the second graph lines closer to the upper right corner are better. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "from hare.visualize import visualize_retrospective_precision_and_recall, visualize_retrospective_roc_curve\n",
    "\n",
    "visualize_retrospective_roc_curve([insult_hare,swear_hare,moba_hare])\n",
    "visualize_retrospective_precision_and_recall([insult_hare,swear_hare,moba_hare])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zooming in on the performance of the machine learning based Hare (the green line) in the precision/recall graph: we see that we can set the toxity threshold really high so we have no false positives, but then we find only 50% of the toxic players (precision 1, recall 0.5)... or we can set the threshold really low so we find all of them, but then 20% are true positives (precision 0.2, recall 1). The optimal threshold seems to have both values are 80%: 80% of toxic players detected, 80% is a true positive."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training\n",
    "======="
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If the existing pretrained models do not score high enough on your conversations, and you have data where the toxic speakers are labeled, you can train your own models. Before you can do this, you need to select a 'brain', which is the classification algorithm HaRe will use under the hood. At the time of writing, neural networks with Bidirectional Gated Recurrent Units are popular for text classification, so we will use the `BiGruBrain`. This 'brain' uses the implementation in TensorFlow."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from hare import Hare\n",
    "from hare.bigrubrain import BiGruBrain\n",
    "\n",
    "custom_hare = Hare()\n",
    "custom_hare.brain = BiGruBrain()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "An interesting option for BiGruBrains (and potentially others) is that they allow you to add word embeddings, which you can see as general knowledge on the meaning of words, collected from large amounts of texts (like Wikipedia)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "custom_hare.brain.embedding_location = '/vol/bigdata/word_embeddings/glove/glove.6B.50d.txt' #This is the path of the embeddings where this notebook was run"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now, all we need to do is add the conversation we want to train on, and we're ready to start training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3\n",
      "5/5 [==============================] - 3s 591ms/step - loss: 0.7564 - acc: 0.2000\n",
      "Epoch 2/3\n",
      "5/5 [==============================] - 0s 15ms/step - loss: 0.6060 - acc: 0.8000\n",
      "Epoch 3/3\n",
      "5/5 [==============================] - 0s 13ms/step - loss: 0.5112 - acc: 0.8000\n"
     ]
    }
   ],
   "source": [
    "custom_hare.add_conversation(convo)\n",
    "custom_hare.add_conversation(second_convo)\n",
    "\n",
    "custom_hare.train()\n",
    "custom_hare.save('name_of_folder')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And that's all there is to it! If you want to reuse the Hare object later, you can save it and put it in the `pretrained` folder, so you can reload it with the `load_pretrained` function later."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}