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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
|
{
"cells": [
{
"metadata": {},
"cell_type": "markdown",
"source": "# Similarity Comparison for Tests",
"id": "b6d4f9b05f293ec8"
},
{
"metadata": {},
"cell_type": "markdown",
"source": "## Fuzzy string matching",
"id": "87c701ae2678e5db"
},
{
"metadata": {},
"cell_type": "code",
"outputs": [],
"execution_count": null,
"source": "# TODO",
"id": "23c3f92212402a87"
},
{
"metadata": {},
"cell_type": "markdown",
"source": "## Sentence Embeddings",
"id": "275e303877d17c08"
},
{
"cell_type": "code",
"id": "initial_id",
"metadata": {
"collapsed": true,
"ExecuteTime": {
"end_time": "2025-06-30T09:41:56.203596Z",
"start_time": "2025-06-30T09:41:54.449778Z"
}
},
"source": [
"\n",
"from sentence_transformers import SentenceTransformer\n",
"\n",
"# The \"all-MiniLM-L6-v2\" model is used for demonstration\n",
"model = SentenceTransformer(\"all-MiniLM-L6-v2\")\n",
"\n",
"expected = [\"Rangieren\", \"420\", \"Gleis\", \"Abstellgleis\", \"auf Sicht fahren\", \"Signal\"]\n",
"actual = [\"Rangieren\", \"420er\", \"Gleis\", \"Abstellgleis\", \"auf Sicht fahren\", \"Signal\"]\n",
"\n",
"sentences = expected + actual\n",
"\n",
"embeddings = model.encode(sentences)\n",
"\n",
"similarities = model.similarity(embeddings, embeddings)\n",
"for i in range(len(expected)):\n",
" print(f\"Expected: {expected[i]}, Actual: {actual[i]}, Similarity: {similarities[i][i+len(expected)]}\")"
],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Expected: Rangieren, Actual: Rangieren, Similarity: 1.0\n",
"Expected: 420, Actual: 420er, Similarity: 0.8273268938064575\n",
"Expected: Gleis, Actual: Gleis, Similarity: 1.0000003576278687\n",
"Expected: Abstellgleis, Actual: Abstellgleis, Similarity: 1.000000238418579\n",
"Expected: auf Sicht fahren, Actual: auf Sicht fahren, Similarity: 1.0\n",
"Expected: Signal, Actual: Signal, Similarity: 1.0\n"
]
}
],
"execution_count": 3
},
{
"metadata": {},
"cell_type": "markdown",
"source": "## Assesment using LLMs",
"id": "dff194fdcd03c59c"
},
{
"metadata": {
"ExecuteTime": {
"end_time": "2025-06-30T13:07:07.597774Z",
"start_time": "2025-06-30T13:07:07.582024Z"
}
},
"cell_type": "code",
"source": [
"import random\n",
"import numpy as np\n",
"from openai import OpenAI\n",
"from dotenv import load_dotenv\n",
"load_dotenv()\n",
"\n",
"client = OpenAI()\n",
"input_text = \"Rangiere mir bitte mal den 420er von Gleis 3 auf das Abstellgleis. Passt auf, du musst auf Sicht bis zu den Signalen fahren.\"\n",
"expected_terms = [\"Rangieren\", \"420\", \"Abstellgleis\", \"auf Sicht fahren\", \"Signal\"]\n",
"actual_terms = [\"Rangieren\", \"420er\", \"Abstellgleis\", \"auf Sicht fahren\", \"Signal\"]\n",
"\n",
"def run_test(shuffle=False):\n",
" expected = expected_terms.copy()\n",
" actual = actual_terms.copy()\n",
" if shuffle:\n",
" random.shuffle(expected)\n",
" random.shuffle(actual)\n",
" print(f\"Expected: {expected}, Actual: {actual}\")\n",
" response = client.chat.completions.create(\n",
" model=\"gpt-4o-mini\",\n",
" response_format={\"type\": \"text\"},\n",
" temperature=0,\n",
" top_p=0,\n",
" messages=[\n",
" {\n",
" \"role\": \"user\",\n",
" \"content\": \"Bewerte die Ähnlichkeit der Ergebnisse der Term Extraktion. Gegeben ist ein Ausgangstext, \"\n",
" \"aus dem Fachbegriffe extrahiert werden mussten. Der Text ist gegeben. Darunter stehen die erwarteten Begriffe, \"\n",
" \"die extrahiert werden sollten. Zum Schluss stehen die tatsächlich extrahierten Begriffe. \"\n",
" \"Bewerte die Ähnlichkeit der extrahierten Begriffe. \"\n",
" \"Ignoriere die Reihenfolge der Begriffe.\"\n",
" \"Sobald sich ein Begriff grundlegend unterscheidet, beende sofort mit FALSE.\"\n",
" \"Wenn ein erwarteter Begriff gänzlich fehlt, beende sofort mit FALSE.\"\n",
" \"Wenn ein Begriff extrahiert wurde, der sicher kein Fachbegriff ist, beende sofort mit FALSE.\"\n",
" \"Ansonsten Ende sofort mit TRUE.\"\n",
" \"Bewerte die extrahierten Begriffe.\"\n",
" },\n",
" {\n",
" \"role\": \"user\",\n",
" \"content\": f\"\"\"{input_text}\\n\\nErwartete Begriffe: {\", \".join(expected)}\\n\\nTatsächliche Begriffe: {\", \".join(actual)}\"\"\"\n",
" }\n",
" ],\n",
" logprobs=True,\n",
" seed=42,\n",
" store=False,\n",
" top_logprobs=5,\n",
" frequency_penalty=0,\n",
" presence_penalty=0,\n",
" )\n",
" result, logprobs = response.choices[0].message.content, response.choices[0].logprobs\n",
" # Get probabilities for TRUE and FALSE as the last token\n",
" probs = {token.token: float(np.exp(token.logprob)) for token in logprobs.content[0].top_logprobs if token.token == \"TRUE\" or token.token == \"FALSE\"}\n",
" # Normalize the probabilities to only account for TRUE and FALSE\n",
" # This might actually distort the result, as other, maybe more likely tokens are ignored (however, such results can be considered faulty)\n",
" total_end = sum(probs.values())\n",
" normalized_probs = {token: value / total_end for token, value in probs.items()}\n",
" print(f\"Probability for success of test {normalized_probs['TRUE']}\")"
],
"id": "6924f5f7741325d0",
"outputs": [],
"execution_count": 49
},
{
"metadata": {
"ExecuteTime": {
"end_time": "2025-06-30T13:07:13.049204Z",
"start_time": "2025-06-30T13:07:09.535572Z"
}
},
"cell_type": "code",
"source": [
"for i in range(5):\n",
" run_test(shuffle=False)"
],
"id": "235b5091f0936d3e",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Probability for success of test 0.2942149597859341\n",
"Probability for success of test 0.46879062662624377\n",
"Probability for success of test 0.2942149597859341\n",
"Probability for success of test 0.26894142136999516\n",
"Probability for success of test 0.24508501864634824\n"
]
}
],
"execution_count": 50
},
{
"metadata": {
"ExecuteTime": {
"end_time": "2025-06-30T13:07:16.452505Z",
"start_time": "2025-06-30T13:07:14.379517Z"
}
},
"cell_type": "code",
"source": [
"for i in range(5):\n",
" run_test(shuffle=True)"
],
"id": "6e5159dd554e4469",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Expected: ['420', 'auf Sicht fahren', 'Rangieren', 'Abstellgleis', 'Signal'], Actual: ['Abstellgleis', 'Rangieren', 'Signal', '420er', 'auf Sicht fahren']\n",
"Probability for success of test 0.2689414096510109\n",
"Expected: ['Rangieren', 'auf Sicht fahren', '420', 'Signal', 'Abstellgleis'], Actual: ['Rangieren', '420er', 'auf Sicht fahren', 'Abstellgleis', 'Signal']\n",
"Probability for success of test 0.053403330553099\n",
"Expected: ['Rangieren', 'auf Sicht fahren', 'Signal', '420', 'Abstellgleis'], Actual: ['Signal', '420er', 'Abstellgleis', 'Rangieren', 'auf Sicht fahren']\n",
"Probability for success of test 0.9820137910906878\n",
"Expected: ['Abstellgleis', '420', 'auf Sicht fahren', 'Signal', 'Rangieren'], Actual: ['420er', 'auf Sicht fahren', 'Signal', 'Abstellgleis', 'Rangieren']\n",
"Probability for success of test 0.8807970779778824\n",
"Expected: ['Rangieren', 'Signal', '420', 'auf Sicht fahren', 'Abstellgleis'], Actual: ['Rangieren', 'auf Sicht fahren', 'Signal', '420er', 'Abstellgleis']\n",
"Probability for success of test 0.09534946618445304\n"
]
}
],
"execution_count": 51
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 2
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython2",
"version": "2.7.6"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|