ppt_chunker.py

from io import StringIO
from typing import List

import pathos.multiprocessing as mp
from unstructured.partition.pptx import partition_pptx
from unstructured.cleaners.core import clean_ordered_bullets, clean_bullets, clean_non_ascii_chars

from ordered_multimap import OrderedMultiIndexMapWeakRef

WRONG_NOUNS = {"BY", "IN", "ON", "INTO", "A", "AT", "WITH", "FROM", "TO", "AND", "OR", "BUT", "OWN", 'BEHIND', 'THROUGH', 'FIERCE', "'S"}
NON_ENDING_PUNCT = {',', ':', ';', "'", '/', '-'}

def process_chunk(chunk, nlp):
    marked = []
    
    for i in range(len(chunk[1])):
        current = chunk[1][i]
        
        if (type(current) is list) and current[1].isupper() and (current[0] == ('Title' or 'UncategorizedText')):
            tokens = nlp(current[1])
            
            try:
                next_ = chunk[1][i+1]

                if type(next_) is not list:
                    continue
                    
            except IndexError:
                continue
                
            if next_[1].isupper() and (next_[0] == ('Title' or 'UncategorizedText')):
                if (tokens[-1].pos_ in {'SYM', "ADP", 'ADV', 'PART', 'PRON', 'DET', "AUX", 'SCONJ', 'CONJ', "CCONJ"}) or ((tokens[-1].pos_ in {'PROPN', 'NOUN', 'VERB'}) and (str(tokens[-1]) in WRONG_NOUNS)):
                    chunk[1][i+1][1] = current[1] + ' ' + next_[1]
                    marked.append(i)

    for i in marked:
        print(f'DELETE: {chunk[1][i]}')
        del chunk[1][i]

    return chunk

def ppt_chunk(file_like, model):
    import time

    s = time.time()
    elements = partition_pptx(file=file_like)

    e = time.time()
    f = e - s
    print(f'TIME {f}')

    chunks = []
    current_chunk = []
    list_items = set()
    marked = set()

    for i, elem in enumerate(elements):
        if elem.category == "PageBreak":
            if current_chunk or list_items:
                if current_chunk:
                    current_chunk = [elem for elem in current_chunk if elem[1] not in marked]
                    
                if list_items:
                    duplicate = marked.intersection(list_items)
                    if duplicate:
                        list_items = list_items - duplicate

                    current_chunk.append("\n".join(list_items))
                    list_items.clear()

                chunks.append([elem.id, current_chunk])
                current_chunk.clear()
        else:
            if elem.text[-1] in NON_ENDING_PUNCT:
                try:
                    next_ = elements[i+1]
                except IndexError:
                    pass
                elements[i+1].text = elem.text + ' ' + next_.text
                marked.add(elem.text)
                
            if (elem.category == "ListItem") or (elem.category == 'NarrativeText'):
                list_items.add(elem.text)
            else:
                current_chunk.append([elem.category, elem.text])

    sr = time.time()
    for chunk in chunks:
        chunk = process_chunk(chunk, model)

    er = time.time()
    fr = er - sr
    print(f'TIME {fr}')
    #with mp.Pool(mp.cpu_count()) as pool:
        #results = pool.imap(process_chunk, chunks)

    print('PASSED AFTER')

    for chunk in chunks:
        for sub_chunk in enumerate(chunk[1]):
            print(f'MODIFIED TEXT {i} : {sub_chunk}')

        
    weakDict = OrderedMultiIndexMapWeakRef()

    metadata_main_title = ''
    
    for pn, slide in enumerate(ppt.slides):
        if pn < from_page:
            continue
        if pn >= to_page:
            break

        try:
            _ = slide.shapes[0]
        except IndexError:
            continue

        text_shapes = [shape for shape in slide.shapes if shape.has_text_frame]

        if len(text_shapes) == 1:
            metadata_main_title = text_shapes[0].text_frame.text
            continue
            
        print(f'SLIDE TEXT: {slide_text}')
        weakDict.insert(slide_text, metadata_main_title)
    return weakDict

    raise NotImplementedError(
        "file type not supported yet(pptx)")