File size: 6,682 Bytes
c0503d0
 
1abbf5b
 
 
 
 
25323eb
4144a2e
 
 
c0503d0
 
52a3679
bcaad94
52a3679
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c0503d0
 
bcaad94
c0503d0
bcaad94
 
c0503d0
bcaad94
 
c0503d0
bcaad94
c0503d0
bcaad94
 
 
 
 
 
 
 
 
 
 
c0503d0
bcaad94
c0503d0
bcaad94
c0503d0
bcaad94
c0503d0
895706c
c0503d0
bcaad94
c0503d0
895706c
 
 
 
 
 
c0503d0
8c737e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c0503d0
 
 
bcaad94
c0503d0
895706c
2a6420a
895706c
c0503d0
 
 
 
895706c
bcaad94
 
 
 
 
 
 
 
 
cbc89c5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
---
library_name: transformers
license: mit
language:
- en
base_model:
- google-bert/bert-base-uncased
pipeline_tag: text-classification
tags:
- PyTorch
- multi-class-classification
---

This is Bert-base-uncased model fine-tuned for topic classification of therapist remarks in psychotherapeutic contexts. The task is a multi-class classification with the following labels:
```python
id2label = {0: 'Time Up and Future Meetings',
            1: 'Complex Emotions Toward Him',
            2: 'Desires and Disappointments',
            3: 'Personal Growth and Decision-Making',
            4: 'Self-Acceptance and Relationships',
            5: 'Understanding and Confronting Fear',
            6: 'See and Understanding Conversations',
            7: 'Clarifying Meaning and Intent',
            8: 'Desire to Escape and Leave',
            9: 'Uncertainty and Understanding Issues',
            10: 'Open Conversation and Sharing',
            11: 'Exploring Emotional Hurt and Bitterness',
            12: 'Guilt and Self-Blame Dynamics',
            13: 'Dynamics of Meaningful Relationships',
            14: 'Struggles and Desires in Learning',
            15: 'Gender Roles and Relationships',
            16: 'Struggles with Personal Change',
            17: 'Complex Mother-Sibling Relationships',
            18: 'Voices and Perception of Sound',
            19: 'Difficulties and Emotional Burdens',
            20: 'Fear and Reflection on Aging',
            21: 'Emotions of Crying and Tears',
            22: 'Father-Child Relationships and Authority',
            23: 'Possibilities and Potential Outcomes',
            24: 'Inner Struggle and Helplessness',
            25: 'Pursuing Meaningful Personal Goals',
            26: 'Job Anxiety and Self-Reflection',
            27: 'Marriage Anxiety and Dependence',
            28: 'Expressions of Anger and Frustration',
            29: 'Nurturing the inner child',
            30: 'Therapy and Father Relationships',
            31: 'Expressions of Happiness and Joy',
            32: 'Revisiting the Past Together',
            33: 'Drinking Habits and Concerns',
            34: 'Managing and Increasing Energy Levels',
            35: 'Safety and Self-Protection Strategies',
            36: 'Understanding Depression and Its Roots'}
```


## Usage

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("AIPsy/bert-base-therapist-topic-classification-eng")
model = AutoModelForSequenceClassification.from_pretrained("AIPsy/bert-base-therapist-topic-classification-eng")

text = "You know, I mean, it seems like you could just go to work and feel so much better."

encoding = tokenizer(
            text,
            truncation=True,
            padding="max_length",
            return_tensors="pt"
        )
output = model(encoding['input_ids'], encoding['attention_mask']).logits
result = np.argmax(output.detach().numpy(), axis=-1)
print(id2label[result[0]])
'Job Anxiety and Self-Reflection'
```

## Dataset

The source material was the recordings of psychotherapeutic sessions posted on YouTube in the public domain. After conducting speaker diarization and transcription of the recordings 15324 items (sentences) were obtained.

## Recommendations

Given the broad context of issues discussed in psychotherapeutic sessions, the authors believe that this model can be used to analyze human communication in general.

## Metrics

Score metrics of trained model
|Sample|F1 macro|
|:-|:-:|
|Test|0.74|
|Validation|0.77|
|Train|0.97|

F1 score metrics for test sample across categories
|ID|Topic|Precision|Recall|F1|
|:-:|:-|:-:|:-:|:-:|
| 0 | Time Up and Future Meetings |  0.81 | 0.80 | 0.80 |
| 1 | Complex Emotions Toward Him |  0.86 | 0.88 | 0.87 |
| 2 | Desires and Disappointments |  0.71 | 0.73 | 0.72 |
| 3 | Personal Growth and Decision-Making |  0.64 | 0.77 | 0.70 |
| 4 | Self-Acceptance and Relationships |  0.69 | 0.82 | 0.75 |
| 5 | Understanding and Confronting Fear |  0.93 | 0.76 | 0.84 |
| 6 | See and Understanding Conversations |  0.77 | 0.85 | 0.81 |
| 7 | Clarifying Meaning and Intent |  0.81 | 0.76 | 0.79 |
| 8 | Desire to Escape and Leave |  0.68 | 0.60 | 0.64 |
| 9 | Uncertainty and Understanding Issues |  0.81 | 0.81 | 0.81 |
| 10 | Open Conversation and Sharing |  0.79 | 0.77 | 0.78 |
| 11 | Exploring Emotional Hurt and Bitterness |  0.74 | 0.70 | 0.72 |
| 12 | Guilt and Self-Blame Dynamics |  0.80 | 0.69 | 0.74 |
| 13 | Dynamics of Meaningful Relationships |  0.75 | 0.75 | 0.75 |
| 14 | Struggles and Desires in Learning |  0.82 | 0.72 | 0.77 |
| 15 | Gender Roles and Relationships |  0.83 | 0.83 | 0.83 |
| 16 | Struggles with Personal Change |  0.76 | 0.60 | 0.67 |
| 17 | Complex Mother-Sibling Relationships |  0.60 | 0.74 | 0.67 |
| 18 | Voices and Perception of Sound |  0.74 | 0.78 | 0.76 |
| 19 | Difficulties and Emotional Burdens |  0.75 | 0.67 | 0.71 |
| 20 | Fear and Reflection on Aging |  0.62 | 0.64 | 0.63 |
| 21 | Emotions of Crying and Tears |  0.65 | 0.76 | 0.70 |
| 22 | Father-Child Relationships and Authority |  0.93 | 0.80 | 0.86 |
| 23 | Possibilities and Potential Outcomes |  0.72 | 0.64 | 0.68 |
| 24 | Inner Struggle and Helplessness |  0.68 | 0.71 | 0.69 |
| 25 | Pursuing Meaningful Personal Goals |  0.58 | 0.65 | 0.61 |
| 26 | Job Anxiety and Self-Reflection |  0.70 | 0.73 | 0.71 |
| 27 | Marriage Anxiety and Dependence |  0.82 | 0.82 | 0.82 |
| 28 | Expressions of Anger and Frustration |  0.73 | 0.76 | 0.75 |
| 29 | Nurturing the inner child |  0.74 | 0.70 | 0.72 |
| 30 | Therapy and Father Relationships |  0.80 | 0.82 | 0.81 |
| 31 | Expressions of Happiness and Joy |  0.78 | 0.79 | 0.78 |
| 32 | Revisiting the Past Together |  0.78 | 0.60 | 0.68 |
| 33 | Drinking Habits and Concerns |  0.69 | 0.79 | 0.73 |
| 34 | Managing and Increasing Energy Levels |  0.54 | 0.62 | 0.58 |
| 35 | Safety and Self-Protection Strategies |  0.65 | 0.85 | 0.73 |
| 36 | Understanding Depression and Its Roots |  0.61 | 0.67 | 0.64 |



## Citation 

- **Papers:** Vanin, A., Bolshev, V., & Panfilova, A. (2024). Applying LLM and Topic Modelling in Psychotherapeutic Contexts. ArXiv, abs/2412.17449. <https://arxiv.org/abs/2412.17449>
- **Developed by:** @myentity, @VadZhen, @Alek123
- **License:** MIT


**BibTeX:**

```
@misc{vanin2024applyingllmtopicmodelling,
      title={Applying LLM and Topic Modelling in Psychotherapeutic Contexts}, 
      author={Alexander Vanin and Vadim Bolshev and Anastasia Panfilova},
      year={2024},
      eprint={2412.17449},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2412.17449}, 
}
```