テキストの感情分析

1
2
3
4
5
6
7
8

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from transformers import pipeline

sentiment_pipeline = pipeline(
    "sentiment-analysis", model="siebert/sentiment-roberta-large-english"
)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

transcript = (
    "Revenue for the third quarter came in at 91 million, in line with previous guidance. "
    "Earnings were strong with net income reaching 47 million. "
    "However, operating costs increased significantly due to supply chain disruptions. "
    "We remain cautious about the near-term outlook given macroeconomic uncertainty. "
    "Our long-term growth strategy continues to deliver results. "
    "Cash flow generation was robust, enabling continued investment in key projects. "
    "We are concerned about rising interest rates impacting our debt servicing costs. "
    "The board has approved a dividend increase reflecting confidence in future earnings."
)

ts_list = [ts.strip() for ts in transcript.split(".") if len(ts.strip()) > 20]
scores = sentiment_pipeline(ts_list)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

results = pd.DataFrame({
    "text": ts_list,
    "label": [s["label"] for s in scores],
    "score": [round(s["score"], 4) for s in scores],
})
results["signed_score"] = results.apply(
    lambda row: row["score"] if row["label"] == "POSITIVE" else -row["score"],
    axis=1,
)
print(results[["label", "score", "text"]].to_string(index=False))

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

fig, ax = plt.subplots(figsize=(10, 4))
colors = ["#10b981" if s > 0 else "#ef4444" for s in results["signed_score"]]
ax.barh(range(len(results)), results["signed_score"], color=colors)
ax.set_yticks(range(len(results)))
ax.set_yticklabels([t[:50] + "..." if len(t) > 50 else t for t in results["text"]], fontsize=8)
ax.set_xlabel("Sentiment Score（+: Positive, -: Negative）")
ax.set_title("文ごとの感情スコア")
ax.axvline(0, color="#64748b", linewidth=0.8)
ax.grid(axis="x", alpha=0.3)
fig.tight_layout()
plt.show()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

pos_count = (results["label"] == "POSITIVE").sum()
neg_count = (results["label"] == "NEGATIVE").sum()
avg_score = results["signed_score"].mean()

print(f"文数:     {len(results)}")
print(f"Positive: {pos_count} ({pos_count/len(results):.0%})")
print(f"Negative: {neg_count} ({neg_count/len(results):.0%})")
print(f"平均スコア: {avg_score:+.4f}")

# 円グラフで構成比を表示
fig, ax = plt.subplots(figsize=(5, 5))
ax.pie(
    [pos_count, neg_count],
    labels=["Positive", "Negative"],
    colors=["#10b981", "#ef4444"],
    autopct="%1.0f%%",
    startangle=90,
)
ax.set_title("センチメント構成比")
plt.show()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

# 各テキストの平均スコアを比較する例
quarterly_scores = {
    "Q1": 0.42,
    "Q2": 0.28,
    "Q3": avg_score,
    "Q4": 0.51,
}

fig, ax = plt.subplots(figsize=(7, 3.5))
bars = ax.bar(quarterly_scores.keys(), quarterly_scores.values(),
              color=["#10b981" if v > 0 else "#ef4444" for v in quarterly_scores.values()])
ax.set_ylabel("平均センチメントスコア")
ax.set_title("四半期ごとのセンチメント推移")
ax.axhline(0, color="#64748b", linewidth=0.8)
ax.grid(axis="y", alpha=0.3)
fig.tight_layout()
plt.show()