Update app.py
Browse files
app.py
CHANGED
@@ -38,7 +38,6 @@ if "history" not in st.session_state:
|
|
38 |
st.session_state.history = []
|
39 |
|
40 |
# 分类函数
|
41 |
-
|
42 |
def classify_emoji_text(text: str):
|
43 |
prompt = f"输入:{text}\n输出:"
|
44 |
input_ids = emoji_tokenizer(prompt, return_tensors="pt").to(emoji_model.device)
|
@@ -50,9 +49,18 @@ def classify_emoji_text(text: str):
|
|
50 |
result = classifier(translated_text)[0]
|
51 |
label = result["label"]
|
52 |
score = result["score"]
|
53 |
-
reasoning =
|
54 |
-
|
55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
return translated_text, label, score, reasoning
|
57 |
|
58 |
# 主页面:输入与分析共存
|
@@ -119,38 +127,42 @@ if st.session_state.history:
|
|
119 |
radar_fig = px.line_polar(radar_df, r='Score', theta='Category', line_close=True, title="⚠️ Risk Radar by Category")
|
120 |
radar_fig.update_traces(line_color='black')
|
121 |
st.plotly_chart(radar_fig)
|
122 |
-
|
123 |
-
|
124 |
-
# 單詞冒犯性分析模塊
|
125 |
st.markdown("### 🧬 Word-level Offensive Correlation")
|
126 |
|
|
|
127 |
last_translated_text = st.session_state.history[-1]["translated"]
|
128 |
words = last_translated_text.split()
|
129 |
|
|
|
130 |
word_scores = []
|
131 |
for word in words:
|
132 |
try:
|
133 |
-
|
134 |
word_scores.append({
|
135 |
"Word": word,
|
136 |
-
"Label":
|
137 |
-
"Score":
|
138 |
})
|
139 |
-
except:
|
140 |
continue
|
141 |
|
142 |
if word_scores:
|
143 |
word_df = pd.DataFrame(word_scores)
|
144 |
word_df = word_df.sort_values(by="Score", ascending=False).reset_index(drop=True)
|
145 |
|
146 |
-
# 顯示前5個,隱藏表格邊框
|
147 |
max_display = 5
|
148 |
-
|
|
|
149 |
|
150 |
display_df = word_df if show_more else word_df.head(max_display)
|
|
|
151 |
st.markdown(
|
152 |
display_df.to_html(index=False, border=0),
|
153 |
unsafe_allow_html=True
|
154 |
)
|
155 |
else:
|
156 |
-
st.info("❕ No word-level analysis available.")
|
|
|
|
|
|
38 |
st.session_state.history = []
|
39 |
|
40 |
# 分类函数
|
|
|
41 |
def classify_emoji_text(text: str):
|
42 |
prompt = f"输入:{text}\n输出:"
|
43 |
input_ids = emoji_tokenizer(prompt, return_tensors="pt").to(emoji_model.device)
|
|
|
49 |
result = classifier(translated_text)[0]
|
50 |
label = result["label"]
|
51 |
score = result["score"]
|
52 |
+
reasoning = (
|
53 |
+
f"The sentence was flagged as '{label}' due to potentially offensive phrases. "
|
54 |
+
"Consider replacing emotionally charged, ambiguous, or abusive terms."
|
55 |
+
)
|
56 |
+
|
57 |
+
st.session_state.history.append({
|
58 |
+
"text": text,
|
59 |
+
"translated": translated_text,
|
60 |
+
"label": label,
|
61 |
+
"score": score,
|
62 |
+
"reason": reasoning
|
63 |
+
})
|
64 |
return translated_text, label, score, reasoning
|
65 |
|
66 |
# 主页面:输入与分析共存
|
|
|
127 |
radar_fig = px.line_polar(radar_df, r='Score', theta='Category', line_close=True, title="⚠️ Risk Radar by Category")
|
128 |
radar_fig.update_traces(line_color='black')
|
129 |
st.plotly_chart(radar_fig)
|
130 |
+
|
131 |
+
# —— 新增:单词级冒犯性相关性分析 —— #
|
|
|
132 |
st.markdown("### 🧬 Word-level Offensive Correlation")
|
133 |
|
134 |
+
# 取最近一次翻译文本,按空格拆分单词
|
135 |
last_translated_text = st.session_state.history[-1]["translated"]
|
136 |
words = last_translated_text.split()
|
137 |
|
138 |
+
# 对每个单词进行分类并收集分数
|
139 |
word_scores = []
|
140 |
for word in words:
|
141 |
try:
|
142 |
+
res = classifier(word)[0]
|
143 |
word_scores.append({
|
144 |
"Word": word,
|
145 |
+
"Label": res["label"],
|
146 |
+
"Score": res["score"]
|
147 |
})
|
148 |
+
except Exception:
|
149 |
continue
|
150 |
|
151 |
if word_scores:
|
152 |
word_df = pd.DataFrame(word_scores)
|
153 |
word_df = word_df.sort_values(by="Score", ascending=False).reset_index(drop=True)
|
154 |
|
|
|
155 |
max_display = 5
|
156 |
+
# Streamlit 1.22+ 支持 st.toggle,若版本不支持可改用 checkbox
|
157 |
+
show_more = st.toggle("Show more words", value=False)
|
158 |
|
159 |
display_df = word_df if show_more else word_df.head(max_display)
|
160 |
+
# 隐藏边框并渲染 HTML 表格
|
161 |
st.markdown(
|
162 |
display_df.to_html(index=False, border=0),
|
163 |
unsafe_allow_html=True
|
164 |
)
|
165 |
else:
|
166 |
+
st.info("❕ No word-level analysis available.")
|
167 |
+
else:
|
168 |
+
st.info("⚠️ No classification data available yet.")
|