Project design collection (artificial intelligence direction): Help newcomers quickly master skills in actual combat, complete project design upgrades independently, and improve their own hard power (not limited to NLP, knowledge graph, computer vision, etc.): Collect meaningful project design collections to help Newcomers quickly master skills in actual combat, helping users make better use of the CSDN platform, independently complete project design upgrades, and improve their own hard power.
Construction of Causal Event Graph Based on Causal Relationship Knowledge Base
1. Project introduction
The real society is a logical society, and a lot of logic, that is, logical experience, exists in our minds, and these logical experiences cannot be exhausted, and it is obviously impractical to rely on a large number of manual summaries. However, fortunately, humans have expressed this logic in words, which provides the possibility for us to use natural language processing technology to realize the extraction of this causal logic. However, limited by my own technical level, it is currently impossible to apply this high-end deep learning method to causal event extraction. Instead, I construct and summarize causal templates, combine the characteristics of the Chinese language, and build a causal language knowledge base.
This project is an attempt to extract causal events and build a causal knowledge graph.
2. Technical route
The causal event graph technical process follows the following process:
It mainly includes the following steps:
1. Construction of causal knowledge base. The construction of causal knowledge base includes causal conjunction database, result lexicon, causal pattern database and so on.
2,. This includes removing noise from the text, removing non-key information, etc.
3. Causal event extraction. This includes causal pair extraction based on causal pattern libraries.
4. Event representation. This is the core issue in the construction of the entire causal graph, because the event graph is essentially connected, and how to choose an appropriate method (phrase, short sentence, sentence stem) is very important.
5. Event integration. Event fusion is very similar to the entity alignment task in knowledge graph
6. Event storage. Event storage is the last step. Based on business needs, it can be stored in a corresponding database, such as a graph database.
3. Final effect
After going through the above several processes, it can support all kinds of queries, such as finding results for known reasons, finding reasons for known results, etc. This is very useful. In short, with the database, there are many things we can do. The next step is Something out of our minds.
Next, query the causal knowledge base with the following events:
the above figures show similar events (one-degree) in the input given event in the database, and the results of similar events (second-degree nodes).
3.1 Tax evasion incidents
3.2 The US attack on Iraq
3.3 Flood event in Shouguang
4. Code display
Part of the code shows:
'''1由果溯因配套式'''
def ruler1(self, sentence):
'''
conm2:〈[之]所以,因为〉、〈[之]所以,由于〉、 <[之]所以,缘于〉
conm2_model:<Conj>{Effect},<Conj>{Cause}
'''
datas = list()
word_pairs =[['之?所以', '因为'], ['之?所以', '由于'], ['之?所以', '缘于']]
for word in word_pairs:
pattern = re.compile(r'\s?(%s)/[p|c]+\s(.*)(%s)/[p|c]+\s(.*)' % (word[0], word[1]))
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][0] + '-' + result[0][2]
data['cause'] = result[0][3]
data['effect'] = result[0][1]
datas.append(data)
if datas:
return datas[0]
else:
return {}
'''2由因到果配套式'''
def ruler2(self, sentence):
'''
conm1:〈因为,从而〉、〈因为,为此〉、〈既[然],所以〉、〈因为,为此〉、〈由于,为此〉、〈只有|除非,才〉、〈由于,以至[于]>、〈既[然],却>、
〈如果,那么|则〉、<由于,从而〉、<既[然],就〉、〈既[然],因此〉、〈如果,就〉、〈只要,就〉〈因为,所以〉、 <由于,于是〉、〈因为,因此〉、
<由于,故〉、 〈因为,以致[于]〉、〈因为,因而〉、〈由于,因此〉、<因为,于是〉、〈由于,致使〉、〈因为,致使〉、〈由于,以致[于] >
〈因为,故〉、〈因[为],以至[于]>,〈由于,所以〉、〈因为,故而〉、〈由于,因而〉
conm1_model:<Conj>{Cause}, <Conj>{Effect}
'''
datas = list()
word_pairs =[['因为', '从而'], ['因为', '为此'], ['既然?', '所以'],
['因为', '为此'], ['由于', '为此'], ['除非', '才'],
['只有', '才'], ['由于', '以至于?'], ['既然?', '却'],
['如果', '那么'], ['如果', '则'], ['由于', '从而'],
['既然?', '就'], ['既然?', '因此'], ['如果', '就'],
['只要', '就'], ['因为', '所以'], ['由于', '于是'],
['因为', '因此'], ['由于', '故'], ['因为', '以致于?'],
['因为', '以致'], ['因为', '因而'], ['由于', '因此'],
['因为', '于是'], ['由于', '致使'], ['因为', '致使'],
['由于', '以致于?'], ['因为', '故'], ['因为?', '以至于?'],
['由于', '所以'], ['因为', '故而'], ['由于', '因而']]
for word in word_pairs:
pattern = re.compile(r'\s?(%s)/[p|c]+\s(.*)(%s)/[p|c]+\s(.*)' % (word[0], word[1]))
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][0] + '-' + result[0][2]
data['cause'] = result[0][1]
data['effect'] = result[0][3]
datas.append(data)
if datas:
return datas[0]
else:
return {}
'''3由因到果居中式明确'''
def ruler3(self, sentence):
'''
cons2:于是、所以、故、致使、以致[于]、因此、以至[于]、从而、因而
cons2_model:{Cause},<Conj...>{Effect}
'''
pattern = re.compile(r'(.*)[,，]+.*(于是|所以|故|致使|以致于?|因此|以至于?|从而|因而)/[p|c]+\s(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][1]
data['cause'] = result[0][0]
data['effect'] = result[0][2]
return data
'''4由因到果居中式精确'''
def ruler4(self, sentence):
'''
verb1:牵动、导向、使动、导致、勾起、引入、指引、使、予以、产生、促成、造成、引导、造就、促使、酿成、
引发、渗透、促进、引起、诱导、引来、促发、引致、诱发、推进、诱致、推动、招致、影响、致使、滋生、归于、
作用、使得、决定、攸关、令人、引出、浸染、带来、挟带、触发、关系、渗入、诱惑、波及、诱使
verb1_model:{Cause},<Verb|Adverb...>{Effect}
'''
pattern = re.compile(r'(.*)\s+(牵动|已致|导向|使动|导致|勾起|引入|指引|使|予以|产生|促成|造成|引导|造就|促使|酿成|引发|渗透|促进|引起|诱导|引来|促发|引致|诱发|推进|诱致|推动|招致|影响|致使|滋生|归于|作用|使得|决定|攸关|令人|引出|浸染|带来|挟带|触发|关系|渗入|诱惑|波及|诱使)/[d|v]+\s(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][1]
data['cause'] = result[0][0]
data['effect'] = result[0][2]
return data
'''5由因到果前端式模糊'''
def ruler5(self, sentence):
'''
prep:为了、依据、为、按照、因[为]、按、依赖、照、比、凭借、由于
prep_model:<Prep...>{Cause},{Effect}
'''
pattern = re.compile(r'\s?(为了|依据|按照|因为|因|按|依赖|凭借|由于)/[p|c]+\s(.*)[,，]+(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][0]
data['cause'] = result[0][1]
data['effect'] = result[0][2]
return data
'''6由因到果居中式模糊'''
def ruler6(self, sentence):
'''
adverb:以免、以便、为此、才
adverb_model:{Cause},<Verb|Adverb...>{Effect}
'''
pattern = re.compile(r'(.*)(以免|以便|为此|才)\s(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][1]
data['cause'] = result[0][0]
data['effect'] = result[0][2]
return data
'''7由因到果前端式精确'''
def ruler7(self, sentence):
'''
cons1:既[然]、因[为]、如果、由于、只要
cons1_model:<Conj...>{Cause},{Effect}
'''
pattern = re.compile(r'\s?(既然?|因|因为|如果|由于|只要)/[p|c]+\s(.*)[,，]+(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][0]
data['cause'] = result[0][1]
data['effect'] = result[0][2]
return data
'''8由果溯因居中式模糊'''
def ruler8(self, sentence):
'''
3
verb2:根源于、取决、来源于、出于、取决于、缘于、在于、出自、起源于、来自、发源于、发自、源于、根源于、立足[于]
verb2_model:{Effect}<Prep...>{Cause}
'''
pattern = re.compile(r'(.*)(根源于|取决|来源于|出于|取决于|缘于|在于|出自|起源于|来自|发源于|发自|源于|根源于|立足|立足于)/[p|c]+\s(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][1]
data['cause'] = result[0][2]
data['effect'] = result[0][0]
return data
'''9由果溯因居端式精确'''
def ruler9(self, sentence):
'''
cons3:因为、由于
cons3_model:{Effect}<Conj...>{Cause}
'''
pattern = re.compile(r'(.*)是?\s(因为|由于)/[p|c]+\s(.*)')
result = pattern.findall(sentence)
data = dict()
if result:
data['tag'] = result[0][1]
data['cause'] = result[0][2]
data['effect'] = result[0][0]
return data
Summarize
1) Based on the set of rules, it is very practical, but there are many problems, and the rules are more maintained. 2) Event representation must be
carefully thought about.
defect