Add additional analysis

Airline Tweets.ipynb

@@ -139,6 +139,7 @@
    "outputs": [],
    "source": [
     "stop_words = stopwords.words('english')\n",
+    "\n",
     "#add custom stop words\n",
     "stop_words.append('http')\n",
     "stop_words.append('hr')\n",
@@ -206,7 +207,7 @@
    "id": "d5e6d696",
    "metadata": {},
    "source": [
-    "We will begin by running multiple types of models with base parameters to see which run best with our dataset. "
+    "We will begin by running multiple types of models with base parameters to see which run best with our dataset. Because we are particularly interested in classifying negative tweets, we will optimize for recall, as we would prefer to wrongly classify positive tweets as negative than wrongly classify negative tweets as positive and potentially miss issues."
    ]
   },
   {
@@ -873,7 +874,7 @@
    "id": "3a9c1ce6",
    "metadata": {},
    "source": [
-    "SVM SMOTE is the highest performing of the three, so I will use this algorithm wth all the models to ensure I select the best performing model."
+    "All three methods are very close -- I chose SVM SMOTE as the highest performing of the three because it has the most balanced recall performance."
    ]
   },
   {
@@ -1366,6 +1367,14 @@
     "plt.show();"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "8183c26c",
+   "metadata": {},
+   "source": [
+    "We see a lot of similar words: thank, best, great, awesome, amazing are words with positive sentiment, and hold, worst, delayed, cancelled, and rude are common words with negative sentiment."
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "53dadded",
@@ -1379,7 +1388,7 @@
    "id": "d44395ba",
    "metadata": {},
    "source": [
-    "Multinomial Naive Bayes, XGBoost, and CatBoost are the three highest performing algorithms, so we will tune parameters for each of these algorithms to improve performance."
+    "Multinomial Naive Bayes and XGBoost are the two highest performing algorithms, so we will tune parameters for each of these algorithms to improve performance."
    ]
   },
   {
@@ -1532,7 +1541,7 @@
    "id": "e63e232c",
    "metadata": {},
    "source": [
-    "Write some analysis"
+    "Interestingly, the original Naive Bayes model is the best performing, so we will implement that model into production."
    ]
   },
   {
@@ -1548,7 +1557,7 @@
    "id": "b7cd46c3",
    "metadata": {},
    "source": [
-    "In order to put this model into production, we will need to build a function that collects tweets, processes them, and categorizes them."
+    "In order to put this model into production, we will need to build a function to processes and categorize a tweet."
    ]
   },
   {
@@ -1562,7 +1571,7 @@
     "    tweet = process(tweet)\n",
     "    tweet = [tweet]\n",
     "    tweet = vectorizer.transform(tweet)\n",
-    "    print(xgb_gridsearch.best_estimator_.predict(tweet))"
+    "    print(models_smote[2].predict(tweet))"
    ]
   },
   {
@@ -1588,12 +1597,21 @@
    ]
   },
   {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "406ffd6a",
+   "cell_type": "markdown",
+   "id": "07cbb482",
    "metadata": {},
-   "outputs": [],
-   "source": []
+   "source": [
+    "## Next Steps"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "be940254",
+   "metadata": {},
+   "source": [
+    "* Develop workflow for reviewing tweet sentiment\n",
+    "* Use sentiment analysis to improve United customer service."
+   ]
   }
  ],
  "metadata": {