text processing

Predict Product Attributes from Product Listings Part 2 – Pipelines & GridSearch

Further improvement on the Product Attributes Text Classifier

This is part 2 of the extracting attributes from product title with the following improvements or add on.

  1. Creating a more generic text cleaning function.
  2. Adding GridSearch for hyper parameters tuning.

Text Cleaning Function

I created a more generic text cleaning function that can accommodate various text data sets. This can use as a base function for text related problem set. The function, if enabled all options, will be able to perform the following:

  1. Converting all text to lowercase.
  2. Stripping html tags especially if data is scrapped from web.
  3. Replacing accented characters with closest English alphabets/characters.
  4. Removing special characters which includes punctuation. Digits may or may not be excluded depending on context. (Digits are not removed for this data set)
  5. Removing stop-words (simple vs detailed. If detailed, will tokenize words before removal else will use simple word replacement.
  6. Removing extra white spaces and newlines.
  7. Normalize text. This either refer to stemming or lemmatizing.

In this example, we only turn on:

  1. converting text to lowercase
  2. remove special characters (need to keep digits) and white spaces,
  3. do a simple stop words removal.

As mentioned in previous post, it is likely a seller would not include much stop words and will try to keep the title as concise as possible given the limited characters and also to make the title more relevant to search engine. As the text length is not too long, will skip normalizing text to save time.

# Text pre-processing modules
from bs4 import BeautifulSoup
import unidecode
import spacy, en_core_web_sm
nlp = spacy.load('en_core_web_sm', disable=['parser', 'ner'])
from nltk.corpus import stopwords
from nltk.tokenize import word_tokenize
from nltk.stem import PorterStemmer
STOPWORDS = set(stopwords.words('english')) 

# Compile regular expression
SPEC_CHARS_REPLACE_BY_SPACE = re.compile('[/(){}\[\]\|@,;]')
SPEC_CHARS = re.compile(r'[^a-zA-z0-9\s]')
SPEC_CHARS_INCLUDE_DIGITS = re.compile(r'[^a-zA-z\s]')
EXTRA_NEWLINES = re.compile(r'[\r|\n|\r\n]+')

## Functions for text preprocessing, cleaning

def strip_htmltags(text):
    soup = BeautifulSoup(text,"lxml")
    return soup.get_text()

def replace_accented_chars(text):
    return unidecode.unidecode(text)

def stem_text(text):
    ps = PorterStemmer()
    modified_txt = ' '.join([ps.stem(word) for word in text.split()])
    return modified_txt    

def lemmatize(text):
    modified_text = nlp(text)
    return ' '.join([word.lemma_ if word.lemma_ != '-PRON-' else word.text for word in modified_text])

def normalize(text, method='stem'):
    """ Text normalization to generate the root form of the inflected words.
        This is done by either "stem" or "lemmatize" the text as defined by the 'method' arguments.
        Note that using "lemmatize" will take much longer to run compared to "stem".
    """
    if method == 'stem':
        return stem_text(text)
    if method == 'lemmatize':
        return lemmatize(text)
    print('Please choose either "stem" or "lemmatize" method to normalize.')
    return text

def rm_special_chars(text, rm_digits=False):
    # remove & replace below special chars with space
    modified_txt = SPEC_CHARS_REPLACE_BY_SPACE.sub(' ', text)

    # remove rest of special chars, no replacing with space
    if rm_digits:
        return SPEC_CHARS_INCLUDE_DIGITS.sub('', modified_txt)
    else:
        return SPEC_CHARS.sub('', modified_txt)

def rm_extra_newlines_and_whitespace(text):
    # rm extra newlines
    modified_txt =  EXTRA_NEWLINES.sub(' ', text)

    # rm extra whitespaces
    return re.sub(r'\s+', ' ', modified_txt)

def rm_stopwords(text, simple=True):
    """ Remove stopwords using either the simple model with replacement.
        or using nltk.tokenize to split the words and replace each words. This will incur speed penalty.
    """
    if simple:
        return ' '.join(word for word in text.split() if word not in STOPWORDS)
    else:
        tokens = word_tokenize(text)
        tokens = [token.strip() for token in tokens]
        return ' '.join(word for word in tokens if word not in STOPWORDS)

def clean_text(raw_text, strip_html = True, replace_accented = True,
                normalize_text = True, normalize_methd = 'stem',
                remove_special_chars = True, remove_digits = True,
                remove_stopwords = True, rm_stopwords_simple_mode = True):

    """ The combined function for all the various preprocessing method.
        Keyword args:
            strip_html               : Remove html tags.
            replace_accented         : Convert accented characters to closest English characters.
            normalize_text           : Normalize text based on normalize_methd.
            normalize_methd          : "stem" or "lemmatize". Default "stem".
            remove_special_chars     : Remove special chars.
            remove_digits            : Remove digits/numeric as special characters.
            remove_stopwords         : Stopwords removal basedon NLTK corpus.
            rm_stopwords_simple_mode : skip tokenize before stopword removal. Speed up time.
    """

    text = raw_text.lower()

    if strip_html:
        text = strip_htmltags(text)
    if replace_accented:
        text = replace_accented_chars(text)
    if remove_special_chars:
        text = rm_special_chars(text, remove_digits)
    if normalize_text:
        text = normalize(text, normalize_methd)
    if remove_stopwords:
        text = rm_stopwords(text, rm_stopwords_simple_mode)

    text = rm_extra_newlines_and_whitespace(text)  

    return text

Grid Search for Hyper Parameters Tuning

Using pipelines, it is easy to incorporate the sklearn grid search to sweep through the various the hyper parameters and select the best value. Two main parameters tuning are:

  1. ngram range in CountVectorizer:
    • In the first part, we only looking a unigram or single word but there are some attributes that are identified by more than one word alone (eg 4G network, 32GB Memory etc) therefore we will sweep the ngram range to find the optimal range.
    • The larger the ngram range the more feature columns will be generated so it will be more memory consuming.
  2. alpha in SGDClassifier
    • This will affect the regularization term and the learning rate of the training model.

With the ngram range and alpha parameters sweep and the best value selected, we can see quite a significant improvement to the accuracy to all the attribute prediction compared to the first version. Most of the improvement comes from the ngram adjusted to (1,3), meaning account for trigram. This is within expectation as more attributes are described by more than one word.

# Prepare model -- Drop na and keep those with values
def get_X_Y_data(x_col, y_col):
    sub_df =  df[[x_col, y_col]]
    sub_df.head()
    sub_df = sub_df.dropna()
    return sub_df[x_col], sub_df[y_col]

# Model training & GridSearch
def generate_model(X, y, verbose = 1):

    text_vect_pipe = Pipeline([
                            ('vect', CountVectorizer()),
                            ('tfidf', TfidfTransformer())
                            ])

    pred_model = Pipeline([
                ('process', text_vect_pipe),
                ('clf', SGDClassifier(loss='hinge', penalty='l2',alpha=1e-3, random_state=42, max_iter=5, tol=None))
               ])

    parameters = {}
    parameters['process__vect__ngram_range'] = [(0,1),(1,2),(1,3)]
    parameters['clf__loss'] = ["hinge"]
    parameters['clf__alpha'] = [5e-6,1e-5]

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state = 42)

    CV = GridSearchCV(pred_model, parameters)
    CV.fit(X_train, y_train)
    y_pred = CV.predict(X_test)

    print('accuracy %s' % accuracy_score(y_pred, y_test))
    print("=="*18)
    print()
    print("Details of GridSearch")

    if verbose:
        print('Best score and parameter combination = ')
        print(CV.best_score_)
        print(CV.best_params_)
        print()
        print("Grid scores on development set:")
        means = CV.cv_results_['mean_test_score']
        stds = CV.cv_results_['std_test_score']
        for mean, std, params in zip(means, stds, CV.cv_results_['params']):
            print("%0.3f (+/-%0.03f) for %r"
                  % (mean, std * 2, params))
        print("=="*18)
        print()

    return CV

X, y = get_X_Y_data('title1', 'Brand')
brand_model = generate_model(X, y)
print('='*29)

The full script is as below. The text cleaning function takes a large part of the code. Excluding the function, the additional of few lines of code for the grid search and pipeline can can bring a relatively significant accuracy improvement.

Next Actions

So far only text features are considered, the next part we will try adding numeric features to see if further improvement can be made.

See Also

  1. Predict Product Attributes from Product Listing Title — Text Feature Extraction and Classification

 

Predict Product Attributes From Product Listing Title — Text Feature Extraction and Classification

Extracting Attributes from Product Title and Image

This is a National (Singapore) Data Science Challenge organised by Shopee hosted on Kaggle. In the advanced category, the tasks is to extract a list of attributes from each product listing given product title and the accompanied image (a text and a image input). Training sets and full instructions are available in the Kaggle link. This is a short attempt of the problem which include the basic data exploration, data cleaning, feature extraction and classification.

Basic Data Exploration

While the project requirement have 3 main product categories, Beauty, Mobile, & Fashion, I will just focus on the Mobile data set. The two other categories will follow the same approach. For the mobile data set, the requirement is to extract the following attributes such as Brand, Phone Model, Camera, Phone Screen Size, Color Family.  A brief exploration of the training data set observed.

  1. Only title (text) & image (pic) available to predict the several attributes
    of the product.
  2. The attributes are already label-encoded.
  3. There are a lot of missing values particularly like Network Connections etc have more than 80% of data missing. This is quite expected as sellers unlikely to put some of these more obscured attributes in the title description while attributes like Brand and Model should have less missing data.

From seller’s perspective, seller will try to include as much information as possible in a
concise manner especially attributes like brands, models etc to make their posting relevant to search and stand out to the buyers. Using only image to extract attributes such as Brand and model might be difficult especially for mobile category where it is difficult to differentiate from pic even with human eye.

From the exploration, I planned the following steps.

  1. Using title (text) as main classification input and ignore images.
  2. Train and predict each attribute at a time.

Basic Data and Text Cleaning

There are some attributes Network Connections, Warranty Period which have large proportion of missing data. However, those attributes have majority of the observations having a certain attribute. In this case, those missing values are assigned with the mode of the training population (e.g. it is likely for Network Connections , most phones should be 4G etc). The attributes are also converted to integer for training purpose.

For the title, before extracting the numeric features, we perform cleaning on the data set. Since most users would highlight the most important feature in the product tile to make their product stand out and relevant, they would generally have omitted most of the stop words, most punctuation. and white spaces Hence for this data set, I will try minimal cleaning: change the title to lowercase and remove special characters. This can reduce a significant amount of time in text cleaning especially for large data set.

Data Cleaning and pipelines

For the advanced data extraction, I chose the Bag-Of-Word (BOW) model to generate the features from the text columns. In the BOW model, I use TF-IDF approach which computes the weighted frequency of each word in each title. For classification, SVM is chosen as the classifier. Pipe-lining makes it easy to streamline the whole text processing and attributes classification making it run on all the different attributes.

Below is the complete code running from extraction, cleaning to classification.

Further Improvement

This is the starting point of the project and take only a few lines of code to get it up and running for quick analysis.  I will improve the existing code by incorporate gridsearch for hyperparameters and expanding on the pipelines and features in the subsequent posts.

See Also

  1. Predict Product Attributes from Product Listings Part 2 – Pipelines & GridSearch

 

Manage and extract data using python and Excel tables

User data such as setting parameters are usually stored in txt, json or perhaps .csv format. One alternative to the common types of storing simple data sets used for initial settings is through using  Excel tables. Excel represents a good way to store and view tables with the extensive formatting options and  different tabs for multiple storage. This provides an edge over txt or other simple data storing. The only difficulty might be that it is not easy to retrieve the data easily unlike in .csv or txt file.

The script below will utilize the Excel tables to extract various information such as setting files, parameters values and at the same time maintain a neat way of viewing and changing all the parameters.

The script will required pyExcel which is interface module between python and excel. The concept is to retrieve all the tables and rows specified within the start and closing tag.

The advantages are:

  1. Multiple tags can be used within the same excel sheet or other excel sheets as well.
  2. Number of columns can be edited easily.
  3. Space can be inserted between rows for easy viewing.
  4. Comment can be inserted so that particular row data can be easily bypassed.
  5. Normal excel formatting can be enabled without disruption to the data retrieved hence allowing easy viewing of data.

Below is a sample table which data will be extracted to be use in subsequent python functions or modules. Note the formatting is up to the user preferences.

sample table

The script run and output are as below. Note that there are various output format to query from. Also notice that the space between the data and those commented rows are being taken care off. The script is available in GitHub.

        xls_set_class = XlsExtractor(fname = r'C:\Python27\Lib\site-packages\excel_table_extract\testset.xls', sheetname= 'Sheet1',
                                     param_start_key = 'start//', param_end_key = 'end//',
                                     header_key = 'header#3//', col_len = 3)
        xls_set_class.open_excel_and_process_block_data()

        print xls_set_class.data_label_list
        ## >>> [u'label1', u'label3']

        print xls_set_class.data_value_list
        ## >>> [[2.0, 3.0], [8.0, 9.0]]

        print xls_set_class.label_value_dict
        ## >>> {u'label1': [2.0, 3.0], u'label3': [8.0, 9.0]}

        print xls_set_class.header_list
        ## >>> [u'header1', u'header2', u'header3']

Python pattern for natural language processing

Python pattern is a good alternative to NLTK with its lightweight and extensive features in natural language processing. In addition, it also have the capability to act as a web crawler and able to retrieve information from twitter, facebook etc. The full functionality can be summarized as stated from their website:

“Pattern is a web mining module for the Python programming language.
It has tools for data mining (Google, Twitter and Wikipedia API, a web crawler, a HTML DOM parser), natural language processing (part-of-speech taggers, n-gram search, sentiment analysis, WordNet), machine learning (vector space model, clustering, SVM), network analysis and <canvas> visualization.”

Below python script illustrate some of the functionality of Python Pattern. I intend to use some of the functions for the google search module developed previously.

The script crawl a particular website, get the plain text of the web page and processed it to remove short sentences (eg links) . After which it will get the top x number of high frequency words found in the web page. After which it will search for all the phrases in the text that contain the high frequency words.

The script still require a number of improvement. For example, keyword ‘turbine’ and ‘turbines’ should be same word and need to classify as one word.

import sys, os, time
from pattern.en import parse, Sentence, parsetree, tokenize
from pattern.search import search
from pattern.vector import count, words, PORTER, LEMMA, Document
from pattern.web import URL, plaintext

def get_plain_text_fr_website(web_address):
    """ Scrape plain text from a web site.
        Args:
            web_address (str): web http address.
        Returns:
            (str): plain text in str.
    """
    s = URL(web_address).download()
    ## s is html format.
    return convert_html_to_plaintext(s)

def convert_html_to_plaintext(html):
    """ Take in html and output as text.
        Args:
            html (str): str in html format.
        Returns:
            (str): plain text in str.

        TODO: include more parameters.
    """
    return plaintext(html)

def retain_text_with_min_sentences_len(raw_text,len_limit =6 ):
    """ Return paragraph with sentences having certain number of length limit.
        Args:
            raw_text (str): text input in paragraphs.
            len_limit (int): min word limit.
        Returns:
            (str): modified text with min words in sentence
    """
    sentence_list  = get_sentences_with_min_words(split_text_to_list_of_sentences(raw_text), len_limit)
    return ''.join(sentence_list)

def split_text_to_list_of_sentences(raw_text):
    """ Split the raw text into list of sentences.
        Args:
            raw_text (str): text input in paragraphs.
        Returns:
            (list): list of str of sentences.
    """
    return tokenize(raw_text)

def get_sentences_with_min_words(sentences_list, len_limit):
    """ Return list of sentences with number of words greater than specified len_limit.
        Args:
            sentences_list (list): sentences break into list.
            len_limit (int): min word limit.
        Returns:
            (list): list of sentences with min num of words.

    """
    return [n for n in sentences_list if word_cnt_in_sent(n) >= len_limit]

def word_cnt_in_sent(sentence):
    """ Return number of words in a sentence. Use spacing as relative word count.
        Count number of alphanum words after splitting the space.
        Args:
            sentence (str): Proper sentence. Can be split from the tokenize function.
        Returns:
            (int): number of words in sentence.
    """
    return len([ n for n in sentence.split(' ') if n.isalnum()]) + 1

def retrieve_string(match_grp):
    """ Function to retrieve the string from the pattern.search.Match class
        Args:
            match_grp (pattern.search.Match): match group
        Returns:
            (str): str containing the words that match
        Note:
            Does not have the grouping selector
    """
    return match_grp.group(0).string

def get_top_freq_words_in_text(txt_string, top_count, filter_method = lambda w: w.lstrip("\'").isalnum(),exclude_len = 0):
    """ Method to get the top frequency of words in text.
        Args:
            txt_string (str): Input string.
            top_count (int): number of top words to be returned.

        Kwargs:
            filter_method (method): special character to ignore, in some cases numbers may also need to ignore.
                                    pass in lambda function.
                                    Default accept method that include only alphanumeric

            exclude_len (int): exclude keyword if len less than certain len.
                                default 0, which will not take effect.

        Returns:
            (list): list of top words """
    docu = Document(txt_string, threshold=1, filter = filter_method)

    ## Provide extra buffer if there is word exclusion
    freq_keyword_tuples = docu.keywords(top=top_count )
    
    ## encode for unicode handliing
    if exclude_len  == 0:
        return [n[1].encode() for n in freq_keyword_tuples]
    else:
        return [n[1].encode() for n in freq_keyword_tuples if not len(n[1])<=exclude_len]

def get_phrases_contain_keyword(text_parsetree, keyword, print_output = 0, phrases_num_limit =5):
    """ Method to return phrases in target text containing the keyword. The keyword is taken as an Noun or NN|NP|NNS.
        The phrases will be a noun phrases ie NP chunks.
        Args:
            text_parsetree (pattern.text.tree.Text): parsed tree of orginal text
            keyword (str): can be a series of words separated by | eg "cat|dog"

        Kwargs:
            print_output (bool): 1 - print the results else do not print.
            phrases_num_limit (int): return  the max number of phrases. if 0, return all.
        
        Returns:
            (list): list of the found phrases. (remove duplication )

        TODO:
            provide limit to each keyword.
    """
    ## Regular expression matching.
    ## interested in phrases containing the traget word, assume target noun is either adj or noun
    target_search_str = 'JJ|NN|NNP|NNS?+ ' + keyword + ' NN|NNP|NNS?+'
    target_search = search(target_search_str, text_parsetree)# only apply if the keyword is top freq:'JJ?+ NN NN|NNP|NNS+'

    target_word_list = []
    for n in target_search:
        if print_output: print retrieve_string(n)
        target_word_list.append(retrieve_string(n))

    target_word_list_rm_duplicates = rm_duplicate_keywords(target_word_list)

    if (len(target_word_list_rm_duplicates)>= phrases_num_limit and phrases_num_limit>0):
        return target_word_list_rm_duplicates[:phrases_num_limit]
    else:
        return target_word_list_rm_duplicates

def rm_duplicate_keywords(target_wordlist):
    """ Method to remove duplication in the key word.
        Args:
            target_wordlist (list): list of keyword str.

        Returns:
            (list): list of keywords with duplicaton removed.
    """
    return list(set(target_wordlist))

if __name__ == '__main__':

    ## random web site for extraction.
    web_address = 'http://en.wikipedia.org/wiki/Turbine'

    ## extract the plain text.
    webtext = get_plain_text_fr_website(web_address)

    ## modified plain text so that it can remove those very short sentences (such as side bar menu).
    modifed_text = retain_text_with_min_sentences_len(webtext)

    ## Begin summarizing the important pt of the website.
    ## first step to get the top freq words, here stated 10.
    ## Exclude len will remove any length less than specified, here stated 2.
    list_of_top_freq_words = get_top_freq_words_in_text(modifed_text, 4, lambda w: w.lstrip("'").isalpha(),exclude_len = 2)
    print list_of_top_freq_words
    ## >> ['turbine', 'turbines', 'fluid', 'impulse']

    ## Parse the whole document for analyzing
    ## The pattern.en parser groups words that belong together into chunks.
    ##For example, the black cat is one chunk, tagged NP (i.e., a noun phrase)
    t = parsetree(modifed_text, lemmata=True)

    ## get target search phrases based on the top freq words.
    for n in list_of_top_freq_words:
        print 'keywords: ', n
        print get_phrases_contain_keyword(t, n)
        print '*'*8

    ##>> keywords:  turbine
    ##>> [u'the Francis Turbine', u'the marine turbine', u'most turbines', u'impulse turbines .Reaction turbines', u'turbine']
    ##>> ********
    ##>> keywords:  turbines
    ##>> [u'de Laval turbines', u'possible .Wind turbines', u'type .Very high efficiency steam turbines', u'conventional steam turbines', u'draft tube .Francis turbines']
    ##>> ********
    ##>> keywords:  fluid
    ##>> [u'a fluid', u'working fluid', u'a high velocity fluid', u'fluid', u'calculations further .Computational fluid']
    ##>> ********
    ##>> keywords:  impulse
    ##>> [u'equivalent impulse', u'impulse', u'Pressure compound multistage impulse', u'de Laval type impulse', u'traditionally more impulse']
    ##>> ********

</pre>
<pre>

Parsing Dict object from text file (More…)

I have modified the DictParser ,mentioned in previous blog, to handle object parsing. Previous version of DictParser can only handle basic data type, whereas in this version, user can pass a dict of objects for the DictParser to identify and it will replace those variables marked with ‘@’, treating them as objects.

An illustration is as below. Note the “second” key has an object @a included in the value list. This will be subsequently substitute by [1,3,4] after parsing.

## Text file
$first
aa:bbb,cccc,1,2,3
1:1,bbb,cccc,1,2,3

$second
ee:bbb,cccc,1,2,3
2:1,bbb,@a,1,2,3  
## end of file

The output from DictParser are as followed:

p = DictParser(temp_working_file, {'a':[1,3,4]}) #pass in a dict with obj def
p.parse_the_full_dict()
print p.dict_of_dict_obj
>>> {'second': {'ee': ['bbb', 'cccc', 1, 2, 3], 2: [1, 'bbb', [1, 3, 4], 1, 2, 3]},
'first': {'aa': ['bbb', 'cccc', 1, 2, 3], 1: [1, 'bbb', 'cccc', 1, 2, 3]}}

If the object is not available or not pass to DictParser, it will be treated as string.

Using the ‘@’ to denote the object is inspired by the Julia programming language where $xxx is used to substitute objects during printing.

Parsing Dict object from text file (Updates)

I have been using the DictParser created as mentioned in previous blog in a recent project to create a setting file for various users. In the project, different users need to have different settings such as parameter filepath.

The setting file created will use the computer name to segregate the different users. By creating a text file (with Dict Parser) based on the different computer names, it is easy to get separate setting parameters for different users. Sample of the setting file are as below.

## Text file
$USER1_COM_NAME
#setting_comment_out:r'c:\data\temp\bbb.txt'
setting2:r'c:\data\temp\ccc.txt'

$USER2_COM_NAME
setting:r'c:\data\temp\eee.txt'
2:1,bbb,cccc,1,2,3
## end of file

The output from DictParser are as followed:

## python output as one dict containing two dicts with different user'USER1_COM_NAME' and 'USER2_COM_NAME'
>> {'USER1_COM_NAME': {'setting2': ['c:\\data\\temp\\ccc.txt']}, 'USER2_COM_NAME': {2: [1, 'bbb', 'cccc', 1, 2, 3], 'setting': ['c:\\data\\temp\\eee.txt']}}

User can use the command “os.environ[‘ComputerName’]” to get the corresponding setting filepath.

I realized that the output format is somewhat similar to json format. This parser is more restrictive in uses hence has some advantage over json in less punctuations (‘{‘, ‘\’) etc and able to comment out certain lines.

Extracting portions of text from text file

I was trying to read the full book of abstracts from a conference earlier and finding it tedious to copy portions of desired paragraphs for my summary report to be fed into my simple auto-summarized module.

I came up with the following script that allows users to put a specific symbol such as “@” at the start and end of the paragraph to mark those paragraphs or sentences to be extracted. More than one portion can be selected and they can be returned as a list for further processing. For my case, each of the paragraph outputted will be auto summarized.

The following diagram illustrated the two different kinds of extraction.

Illustration of extraction type

The script scans all the lines of the text file, looking for the key_symbol (“@” in this case) and marks the index of the selected lines. The present method only use string “startwith” function. It can be expanded to be using regular expression.

Depending on the mode (overlapping or non-overlapping), it will calculate the portion of the text to be selected and output as a list which can be use for further processing.

Script can be found here.

 

Scaping google results using python (Part 3)

The  post on the testing of google search script I created last week describe the limitations of the script to scrape the required information. The search phrase is “best hotels to stay in Tokyo”. My objective is to find suitable and popular hotels to stay in Tokyo and within the budget limit.

The other limitation is that the script can only take in one input or key phrase at one go. This is not very useful. Users would tend to search a variation of the key phrases to get the desirable results. I done some modifications to the script so it can take in either a key phrase (str) or  a list of key phrases (list) so it can search all the key phrases at one go.

The script will now iterate the search phrases. Below is the summarized flow:

  1. For each key phrase in key phrase list, generate the associated google search url, append all url to list.
  2. For the list of google search url, Scrapy will scrape the individual url for the google results links. Append all links to a output file. There is one drawback. The links for the first key phrases will be displayed first followed by the 2nd key phrase.
  3. For each of the links, Scrapy will scrape the content namely the title, meta description and for now, if specified,  all the text within the <p> tag.
  4. The resulting file will be very big depending on the size of the search results.

The format of the output is still not to satisfaction. Also printing all the <p> tag does not accomplished much in summarizing what I need.

The next step, hopefully, can utilize some of the NLTK and summarize tools to help filter the results.

The current script is in Git Hub.

Getting Google Search results with python (testing the program)

I was testing out the google search script I created last week. I was searching for the “best hotels to stay in Tokyo”. My objective is to find suitable and popular hotels to stay in Tokyo and within the budget limit.

The python module was created with the intention to display more meaningful and relevant data without clicking to individual websites. However, with just the meta title and meta contents from the search results, it is not really useful in obtaining meaningful results.

I tried to modify the module by extraction of the paragraphs from each site and output them together with the meta descriptions. I make some changes to the script to handle  multiple newline characters and debug on the unicode error that keeps popping out when output the text results.

To extract the paragraphs from each site, I used the xpath command as below.

sel = Selector(response)
paragraph_list = sel.xpath('//p/text()').extract()

To handle the unicode identification error, the following changes are made. The stackoverflow link provides the solution to the problem.

## convert the paragraph list to one continuous string
para_str = self.join_list_of_str(paragraph_list, joined_chars= '..')
## Replace any unknown unicode characters with ?
para_str = para_str.encode(errors='replace')
## Remove newline characters
para_str = self.remove_whitespace_fr_raw(para_str)

With the paragraphs displayed at the output, I was basically reading large chunks of texts and it was certainly messy with the newline removed. I could not really get good information out of it.

For example, it is better to get the ranked hotels from tripadvisor site but from the google search module, tripadvisor only displays the top page without any hotels listed. Below is the output I get from TripAdvisor site pertaining to the search result.

Tokyo Hotels: Check Out 653 hotels with 77,018 Reviews – TripAdvisor
ttp://www.tripadvisor.com.sg/Hotels-g298184-Tokyo_Tokyo_Prefecture_Kanto-Hotels.html

Tokyo Hotels: Find 77,018 traveller reviews and 2,802 candid photos for 653 hotels in Tokyo, Japan on TripAdvisor.

Price per night..Property type..Neighbourhood..Traveller rating..Hotel class..Amenities..Property name..Hotel brand

Performing recursive crawling on TripAdvisor itself perhaps will achieve more meaningful results.

Currently, I do not have much idea on enhancing the script to extract more meaningful data. Perhaps I can use text processing to summarize the paragraphs into meaningful data which would be the next step, utilizing the NLTK module. However, I am not hopeful of the final results.

For this particular search query, perhaps it would be easier to cater specific crawling methods on several target website such as TripAdvisor, Agoda etc rather than a general extraction of text.