BIDI_ML_INFO_EXTRACTION/BiddingKG/deepdive/udf/supervise_guest_WinnerOfBidding.py

#!/usr/bin/env python
#encoding:utf-8
from deepdive import *
import random
from collections import namedtuple
from commonutil import *
import re


Label = namedtuple('Label', 'entity_id, label, type')

@tsv_extractor
@returns(lambda
        entity_id   = "text",
        label   = "int",
        rule_id = "text",
    :[])
# heuristic rules for finding positive/negative examples of transaction relationship mentions
def supervise(
        entity_id="text", entity_begin="int", entity_end="int",
        doc_id="text", sentence_index="int", sentence_text="text",
        tokens="text[]",  pos_tags="text[]", ner_tags="text[]",
    ):

    # Constants
    label = Label(entity_id=entity_id,label=None,type=None)
    Bidding = frozenset(["中标","中标人","中"])
    LOCA = frozenset(["单位","人","中标人","标人"])
    MAX_DIST = 15
    TYPE_MENTION = frozenset(["org","company"])

    # Common data objects
    if entity_begin>MAX_DIST:
        begin = entity_begin-MAX_DIST
    else:
        begin = 0
    if len(tokens)-entity_end>MAX_DIST:
        end = entity_end+MAX_DIST
    else:
        end = -1
    front_tokens = tokens[begin:entity_begin]
    end_tokens = tokens[entity_end:end]
    front_ner = ner_tags[begin:entity_begin]
    end_ner = ner_tags[end:entity_end]
    
    log(sentence_text)
    

    # Rule: Candidates that are too far apart
    patten = "(中标[人|单位|机构|])"
    sear = re.search(patten,"".join(front_tokens))
    if sear:
        if len(TYPE_MENTION.intersection(front_ner[get_index_in_tokens(sear.end(0),front_tokens):]))>0:
            yield label._replace(label=-1,type="pos:third entity between")
        else:
            yield label._replace(label=1,type="中标单位:entity")
    else:
        yield label._replace(label=-1,type="pos:no match")
            
        
    '''
    if len(LOCA.intersection(front_tokens))>0:
        if len(Bidding.intersection(front_tokens))>0:
            comb = combine(list(Bidding),list(LOCA))
            index = find_index(comb,"".join(front_tokens))
            if index>0:
                if len(TYPE_MENTION.intersection(front_ner[index:entity_begin]))>0:
                    yield label._replace(label=-1,type="pos:third entity between")
                else:
                    yield label._replace(label=1,type="中标单位")
            else:
                yield label._replace(label=-1,type="pos:not match all")
        else:
            yield label._replace(label=-1,type="pos:no match 单位")
    else:
        yield label._replace(label=-1,type="pos:no match")
    '''