Update app.py

app.py

@@ -1,6 +1,12 @@
 import pkg_resources
-from transformers import pipeline
+from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
 import gradio as gr
+from ferret import Benchmark
+
+sentiment_tokenizer = AutoTokenizer.from_pretrained("Birkir/electra-base-igc-is-sentiment-analysis")
+sentiment_model = AutoModelForSequenceClassification.from_pretrained("Birkir/electra-base-igc-is-sentiment-analysis")
+
+sentiment_bench = Benchmark(sentiment_model, sentiment_tokenizer)
 
 translator = pipeline("translation", model="Helsinki-NLP/opus-mt-is-en")
 sentiment_classifier = pipeline("text-classification", model="Birkir/electra-base-igc-is-sentiment-analysis")
@@ -8,6 +14,31 @@ formality_classifier = pipeline("text-classification", model="svanhvit/formality
 detoxify_pipeline = pipeline('text-classification', model='unitary/toxic-bert', tokenizer='bert-base-uncased', function_to_apply='sigmoid', top_k=None)
 politeness_classifier = pipeline("text-classification", model="Genius1237/xlm-roberta-large-tydip")
 
+def analyze_sentiment_with_influence(icelandic_text):
+    sentiment_label, sentiment_score = analyze_sentiment(icelandic_text)
+    
+    # Convert sentiment label
+    sentiment_label = sentiment_label.replace("LABEL_", "")
+    
+    # Use Ferret to analyze influential words for sentiment
+    explanations_sentiment = sentiment_bench.explain(icelandic_text, target=1)  # Adjust target as necessary
+    
+    influential_words = []
+    for explanation in explanations_sentiment:
+        if explanation.explainer == 'Partition SHAP':
+            tokens = replace_encoding(explanation.tokens)
+            token_score_pairs = zip(tokens, explanation.scores)
+            influential_words.extend([(token, score) for token, score in token_score_pairs])
+    
+    # Format your response to include influential words
+    influential_words_str = "; ".join([f"{token} ({score:.2f})" for token, score in influential_words])
+    
+    analysis_results = (
+        f"Sentiment: Label: {sentiment_label}, Score: {round(sentiment_score, 2)}\n"
+        f"Influential Words: {influential_words_str}"
+    )
+    return analysis_results
+
 def translate_text(text):
     translation = translator(text, max_length=512)
     return translation[0]['translation_text']