EMO355 (Jerteh-355 Emotion & Evocation Embeddings)

Usage / Upotreba

Model se u potpunosti učitava i koristi preko sentence_transformers biblioteke:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("procesaur/Emo355")
test_sentences = [
        "Nisam siguran šta će se desiti sutra, ali se nadam najboljem."
        "Ovo je nedopustivo! Kakav užasan uspeh i sramota za celu zemlju.",
    ]

# Get 16-dimensional emotion vector predictions directly
predictions = model.encode(test_sentences)

LABELS = [
    # 8 Base
    "anger", "anticipation", "disgust", "fear",
    "joy", "sadness", "surprise", "trust",
    # 8 Evocation
    "evoc_anger", "evoc_anticipation", "evoc_disgust", "evoc_fear",
    "evoc_joy", "evoc_sadness", "evoc_surprise", "evoc_trust"
]

for text, pred in zip(test_sentences, predictions):
    print(f"\nText: {text}")
    scores = dict(zip(LABELS, pred.tolist()))
    for k, v in scores.items():
        print(f" {k:<20}: {v:.4f}")

Text: Nisam siguran šta će se desiti sutra, ali se nadam najboljem.
 anger               : 0.0026
 anticipation        : 0.4779
 disgust             : 0.0189
 fear                : 0.0056
 joy                 : 0.2393
 sadness             : 0.1392
 surprise            : 0.1545
 trust               : 0.0844
 evoc_anger          : 0.2122
 evoc_anticipation   : 0.2360
 evoc_disgust        : -0.0765
 evoc_fear           : 0.0729
 evoc_joy            : 0.3163
 evoc_sadness        : 0.1806
 evoc_surprise       : 0.1333
 evoc_trust          : 0.0023

Text: Ovo je nedopustivo! Kakav užasan uspeh i sramota za celu zemlju.
 anger               : 0.5147
 anticipation        : 0.0130
 disgust             : 0.1701
 fear                : 0.0393
 joy                 : 0.0718
 sadness             : 0.0931
 surprise            : 0.1342
 trust               : 0.0176
 evoc_anger          : 0.4144
 evoc_anticipation   : 0.0837
 evoc_disgust        : -0.0117
 evoc_fear           : 0.0504
 evoc_joy            : 0.0973
 evoc_sadness        : 0.1900
 evoc_surprise       : 0.1695
 evoc_trust          : 0.0149

import matplotlib.pyplot as plt
import numpy as np
import base64
from io import BytesIO

def plot_radar_chart(embedding, labels, color="blue", title=""):
    plt.rcParams["font.family"] = "Times New Roman"
    num_vars = len(embedding)
    angles = np.linspace(0, 2 * np.pi, num_vars, endpoint=False).tolist()
    angles += angles[:1]  # Complete the loop
    fig, ax = plt.subplots(figsize=(5, 5), subplot_kw=dict(polar=True))

    ax.set_title(title, fontdict={"fontsize": 20, "fontweight": "bold"}, loc="left")

    ax.set_theta_offset(np.pi / 2)
    ax.set_theta_direction(-1)
    ax.set_xticks(angles[:-1])
    plt.xticks(angles[:-1], labels, size=20)
    plt.yticks([0.1, 0.5, 1, 2], ["0.1", "0.5", "1", "2"], color="grey", size=10)
    plt.ylim(0, 1)

    embedding.append(embedding[0])
    embedding = [0 if x<0 else x for x in embedding]
    ax.plot(angles, embedding, linewidth=2, linestyle='solid')
    ax.fill(angles, embedding, color=color, alpha=0.25)
    plt.legend(loc='upper right', bbox_to_anchor=(0.1, 0.1))

    buffer = BytesIO()
    fig.savefig(buffer, format="png")
    buffer.seek(0)
    img_base64 = base64.b64encode(buffer.read()).decode("utf-8")
    return f'<img src="data:image/png;base64,{img_base64}" />'

scores = list(predictions[1].squeeze())
print(plot_radar_chart(scores[0:8], LABELS[0:8], "blue", "emotion") + plot_radar_chart(scores[8:16], LABELS[8:16], "green", "evocation"))

Cit.

@inproceedings{vskoric2025embedding,
  title={Embedding Text in Emotion and Evocation Vector Spaces},
  author={{\v{S}}kori{\'c}, Mihailo and Stankovi{\'c}, Ranka},
  booktitle={ReLDI 2025 Synergies},
  pages={63-67},
  year={2025}
}