CoolFace
Apppublic

awacke1/Knowledge-graphs

sourceHugging Faceupdated 4y agoView on Hugging Face
3likes
app.py256 linesDownload Raw Back to root
1from logging import disable
2from pkg_resources import EggMetadata
3import streamlit as st
4import streamlit.components.v1 as components
5import networkx as nx
6import matplotlib.pyplot as plt
7from pyvis.network import Network
8from streamlit.state.session_state import SessionState
9from streamlit.type_util import Key
10import rebel
11import wikipedia
12from utils import clip_text
13from datetime import datetime as dt
14import os
15
16MAX_TOPICS = 3
17
18wiki_state_variables = {
19    'has_run_wiki':False,
20    'wiki_suggestions': [],
21    'wiki_text' : [],
22    'nodes':[],
23    "topics":[],
24    "html_wiki":""
25}
26
27free_text_state_variables = {
28    'has_run_free':False,
29    "html_free":""
30
31}
32
33BUTTON_COLUMS = 4
34
35def wiki_init_state_variables():
36    for k in free_text_state_variables.keys():
37        if k in st.session_state:
38            del st.session_state[k]
39
40    for k, v in wiki_state_variables.items():
41        if k not in st.session_state:
42            st.session_state[k] = v
43
44def wiki_generate_graph():
45    st.session_state["GRAPH_FILENAME"] = str(dt.now().timestamp()*1000) + ".html"
46
47    if 'wiki_text' not in st.session_state:
48        return
49    if len(st.session_state['wiki_text']) == 0:
50        st.error("please enter a topic and select a wiki page first")
51        return
52    with st.spinner(text="Generating graph..."):
53        texts = st.session_state['wiki_text']
54        st.session_state['nodes'] = []
55        nodes = rebel.generate_knowledge_graph(texts, st.session_state["GRAPH_FILENAME"])
56        HtmlFile = open(st.session_state["GRAPH_FILENAME"], 'r', encoding='utf-8')
57        source_code = HtmlFile.read()
58        st.session_state["html_wiki"] = source_code
59        os.remove(st.session_state["GRAPH_FILENAME"])
60        for n in nodes:
61            n = n.lower()
62            if n not in st.session_state['topics']:
63                possible_topics = wikipedia.search(n, results = 2)
64                st.session_state['nodes'].extend(possible_topics)
65        st.session_state['nodes'] = list(set(st.session_state['nodes']))
66        st.session_state['has_run_wiki'] = True
67    st.success('Done!')
68
69def wiki_show_suggestion():
70    st.session_state['wiki_suggestions'] = []
71    with st.spinner(text="fetching wiki topics..."):
72        if st.session_state['input_method'] == "wikipedia":
73            text = st.session_state.text
74            if (text is not None) and (text != ""):
75                subjects = text.split(",")[:MAX_TOPICS]
76                for subj in subjects:
77                    st.session_state['wiki_suggestions'] += wikipedia.search(subj, results = 3)
78
79def wiki_show_text(page_title):
80    with st.spinner(text="fetching wiki page..."):
81        try:
82            page = wikipedia.page(title=page_title, auto_suggest=False)
83            st.session_state['wiki_text'].append(clip_text(page.summary))
84            st.session_state['topics'].append(page_title.lower())
85            st.session_state['wiki_suggestions'].remove(page_title)
86
87        except wikipedia.DisambiguationError as e:
88            with st.spinner(text="Woops, ambigious term, recalculating options..."):
89                st.session_state['wiki_suggestions'].remove(page_title)
90                temp = st.session_state['wiki_suggestions'] + e.options[:3]
91                st.session_state['wiki_suggestions'] = list(set(temp))
92        except wikipedia.WikipediaException:
93            st.session_state['wiki_suggestions'].remove(page_title)
94
95def wiki_add_text(term):
96    if len(st.session_state['wiki_text']) > MAX_TOPICS:
97        return
98    try:
99        page = wikipedia.page(title=term, auto_suggest=False)
100        extra_text = clip_text(page.summary)
101
102        st.session_state['wiki_text'].append(extra_text)
103        st.session_state['topics'].append(term.lower())
104        st.session_state['nodes'].remove(term)
105
106    except wikipedia.DisambiguationError as e:
107        print(e)
108        with st.spinner(text="Woops, ambigious term, recalculating options..."):
109            st.session_state['nodes'].remove(term)
110            temp = st.session_state['nodes'] + e.options[:3]
111            st.session_state['nodes'] = list(set(temp))
112    except wikipedia.WikipediaException as e:
113        print(e)
114        st.session_state['nodes'].remove(term)
115
116def wiki_reset_session():
117    for k in wiki_state_variables:
118        del st.session_state[k]
119
120def free_reset_session():
121    for k in free_text_state_variables:
122        del st.session_state[k]
123
124def free_text_generate():
125    st.session_state["GRAPH_FILENAME"] = str(dt.now().timestamp()*1000) + ".html"
126    text = st.session_state['free_text'][0:100]
127    rebel.generate_knowledge_graph([text], st.session_state["GRAPH_FILENAME"])
128    HtmlFile = open(st.session_state["GRAPH_FILENAME"], 'r', encoding='utf-8')
129    source_code = HtmlFile.read()
130    st.session_state["html_free"] = source_code
131    os.remove(st.session_state["GRAPH_FILENAME"])
132    st.session_state['has_run_free'] = True
133
134def free_text_layout():
135    st.text_area("Free text", key="free_text", height=5, value="Tardigrades, known colloquially as water bears or moss piglets, are a phylum of eight-legged segmented micro-animals.")
136    st.button("Generate", on_click=free_text_generate, key="free_text_generate")
137
138def free_test_init_state_variables():
139    for k in wiki_state_variables.keys():
140        if k in st.session_state:
141            del st.session_state[k]
142
143    for k, v in free_text_state_variables.items():
144        if k not in st.session_state:
145            st.session_state[k] = v
146
147st.title('RE:Belle')
148st.markdown(
149"""
150### Building Beautiful Knowledge Graphs With REBEL
151""")
152st.selectbox(
153     'input method',
154     ('wikipedia', 'free text'),  key="input_method")
155
156
157def show_wiki_hub_page():
158    st.sidebar.button("Reset", on_click=wiki_reset_session, key="reset_key")
159
160    st.sidebar.markdown(
161"""
162## How To Create a Graph:
163- Enter wikipedia search terms, separated by comma's
164- Choose one or more of the suggested topics (max 3)
165- Click generate!
166"""
167)
168    cols = st.columns([8, 1])
169    with cols[0]:
170        st.text_input("wikipedia search term", on_change=wiki_show_suggestion, key="text", value="graphs, are, awesome")
171    with cols[1]:
172        st.text('')
173        st.text('')
174        st.button("Search", on_click=wiki_show_suggestion, key="show_suggestion_key")
175
176    if len(st.session_state['wiki_suggestions']) != 0:
177        num_buttons = len(st.session_state['wiki_suggestions'])
178        num_cols = num_buttons if 0 < num_buttons < BUTTON_COLUMS else BUTTON_COLUMS
179        columns = st.columns([1] * num_cols )
180        for q in range(1 + num_buttons//num_cols):
181            for i, (c, s) in enumerate(zip(columns, st.session_state['wiki_suggestions'][q*num_cols: (q+1)*num_cols])):
182                with c:
183                    st.button(s, on_click=wiki_show_text, args=(s,), key=str(i)+s+"wiki_suggestion")
184
185    if len(st.session_state['wiki_text']) != 0:
186        for i, t in enumerate(st.session_state['wiki_text']):
187            new_expander = st.expander(label=t[:30] + "...", expanded=(i==0))
188            with new_expander:
189                st.markdown(t)
190
191    if len(st.session_state['wiki_text']) > 0:
192        st.button("Generate", on_click=wiki_generate_graph, key="gen_graph")
193    st.sidebar.markdown(
194        """
195    ## How to expand the graph
196    - Click a button below the graph to expand that node
197     (Only nodes that have wiki pages will be expanded)
198    - Hit the Generate button again to expand your graph!
199    """
200    )
201
202    if st.session_state['has_run_wiki']:
203
204        components.html(st.session_state["html_wiki"], width=720, height=600)
205        num_buttons = len(st.session_state["nodes"])
206        num_cols = num_buttons if 0 < num_buttons < BUTTON_COLUMS else BUTTON_COLUMS
207        columns = st.columns([1] * num_cols + [1])
208
209        for q in range(1 + num_buttons//num_cols):
210            for i, (c, s) in enumerate(zip(columns, st.session_state["nodes"][q*num_cols: (q+1)*num_cols])):
211                with c:
212                    st.button(s, on_click=wiki_add_text, args=(s,), key=str(i)+s)
213
214def show_free_text_hub_page():
215    st.sidebar.button("Reset", on_click=free_reset_session, key="free_reset_key")
216    st.sidebar.markdown(
217"""
218## How To Create a Graph:
219- Enter a text you'd like to see as a graph.
220- Click generate!
221"""
222)
223
224    free_text_layout()
225
226    if st.session_state['has_run_free']:
227        components.html(st.session_state["html_free"], width=720, height=600)
228
229if st.session_state['input_method'] == "wikipedia":
230    wiki_init_state_variables()
231    show_wiki_hub_page()
232else:
233    free_test_init_state_variables()
234    show_free_text_hub_page()
235
236
237
238st.sidebar.markdown(
239"""
240## What This Is And Why We Built it
241
242This space shows how a transformer network can be used to convert *human* text into a computer-queryable format: a **knowledge graph**. Knowledge graphs are graphs where each node (or *vertex* if you're fancy) represent a concept/person/thing and each edge the link between those concepts. If you'd like to know more, you can read [this blogpost](https://www.ml6.eu/knowhow/knowledge-graphs-an-introduction-and-business-applications).
243
244Knowledge graphs aren't just cool to look at, they are an extremely versatile way of storing data, and are used in machine learning to perform tasks like fraud detection. You can read more about the applications of knowledge graphs in ML in [this blogpost](https://blog.ml6.eu/how-are-knowledge-graphs-and-machine-learning-related-ff6f5c1760b5).
245
246There is one problem though: building knowledge graphs from scratch is a time-consuming and tedious task, so it would be a lot easier if we could leverage machine learning to **create** them from existing texts. This demo shows how a model named **REBEL** has been trained to do just that: it reads summaries from Wikipedia (or any other text you input), and generates a graph containing the information it distills from the text.
247"""
248)
249
250st.sidebar.markdown(
251"""
252*Credits for the REBEL model go out to Pere-Lluís Huguet Cabot and Roberto Navigli.
253The code can be found [here](https://github.com/Babelscape/rebel),
254and the original paper [here](https://github.com/Babelscape/rebel/blob/main/docs/EMNLP_2021_REBEL__Camera_Ready_.pdf)*
255"""
256)