Update README.md

README.md

@@ -18,13 +18,27 @@ should probably proofread and complete it, then remove this comment. -->
 
 # bert-base-finance-sentiment-noisy-search
 
-This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on the None dataset.
-It achieves the following results on the evaluation set:
-- Loss: 0.8976
-- F1: 0.8595
-- Accuracy: 0.8592
-- Precision: 0.8600
-- Recall: 0.8592
+This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on Kaggle finance news sentiment analysis with data enhancement using noisy search. The process is explained below:
+
+1. First "bert-base-uncased" was fine-tuned on Kaggle's finance news sentiment analysis https://www.kaggle.com/ankurzing/sentiment-analysis-for-financial-news dataset achieving accuracy of about 88%
+2. We then used a logistic-regression classifier on the same data. Here we looked at coefficients that contributed the most to the "Positive" and "Negative" classes by inspecting only bi-grams. 
+3. Using the top 25 bi-grams per class (i.e. "Positive" / "Negative") we invoked Bing news search with those bi-grams and retrieved up to 50 news per bi-gram phrase.
+4. We called it "noisy-search" because it is assumed the positive bi-grams (e.g. "profit rose" , "growth net") give rise to positive examples whereas negative bi-grams (e.g. "loss increase", "share loss") result in negative examples.
+5. For each article we kept the title + excerpt and labeled it according to the noisy pre-assumption
+6. We then trained the same model on the noisy data and apply it to an held-out test set from the original set split.
+7. Training with couple of thousands noisy "positives" and "negatives" yielded a test set accuracy of about 95%.
+
+Accuracy results for Logistic Regression (LR) and BERT (base-cased) are shown below
+
+Accuracy |  Kaggle    |  Enhanced noisy data set
+LR       |   79.0%    |       95.1%
+BERT     |   88.7%    |       95.2%
+
+Here we describe the process in more detail alongside with other metrics -  https://drive.google.com/file/d/1MI9gRdppactVZ_XvhCwvoaOV1aRfprrd/view?usp=sharing 
+
+
+
+
 
 ## Model description
 
@@ -51,16 +65,6 @@ The following hyperparameters were used during training:
 - lr_scheduler_type: linear
 - num_epochs: 5
 
-### Training results
-
-| Training Loss | Epoch | Step | Validation Loss | F1     | Accuracy | Precision | Recall |
-|:-------------:|:-----:|:----:|:---------------:|:------:|:--------:|:---------:|:------:|
-| 0.4782        | 1.0   | 1022 | 0.4979          | 0.8357 | 0.8370   | 0.8376    | 0.8370 |
-| 0.3233        | 2.0   | 2044 | 0.5273          | 0.8568 | 0.8571   | 0.8568    | 0.8571 |
-| 0.1874        | 3.0   | 3066 | 0.6467          | 0.8531 | 0.8533   | 0.8534    | 0.8533 |
-| 0.1048        | 4.0   | 4088 | 0.7924          | 0.8600 | 0.8598   | 0.8608    | 0.8598 |
-| 0.0556        | 5.0   | 5110 | 0.8976          | 0.8595 | 0.8592   | 0.8600    | 0.8592 |
-
 
 ### Framework versions