Semantic Search with Approximate Nearest Neighbors and Text Embeddings from TF-Hub

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

This tutorial illustrates how to generate embeddings from a TensorFlow Hub (TF-Hub) module given input data, and build an approximate nearest neighbours (ANN) index using the extracted embeddings. The index can then be used for real-time similarity matching and retrieval.

When dealing with a large corpus of data, it's not efficient to perform exact matching by scanning the whole repository to find the most similar items to a given query in real-time. Thus, we use an approximate similarity matching algorithm which allows us to trade off a little bit of accuracy in finding exact nearest neighbor matches for a significant boost in speed.

In this tutorial, we show an example of real-time text search over a corpus of news headlines to find the headlines that are most similar to a query. Unlike keyword search, this captures the semantic similarity encoded in the text embedding.

The steps of this tutorial are:

  1. Download sample data.
  2. Generate embeddings for the data using a TF-Hub module
  3. Build an ANN index for the embeddings
  4. Use the index for similarity matching

We use Apache Beam with TensorFlow Transform (TF-Transform) to generate the embeddings from the TF-Hub module. We also use Spotify's ANNOY library to build the approximate nearest neighbours index. You cam find benchmarking of ANN framework in this Github repository.

This tutorial uses TensorFlow 1.0 and works only with TF1 Hub modules from TF-Hub. See the updated TF2 version of this tutorial.

Setup

Install the required libraries.

pip install -q tensorflow_transform
pip install -q apache_beam
pip install -q sklearn
pip install -q annoy

Import the required libraries

import os
import sys
import pathlib
import pickle
from collections import namedtuple
from datetime import datetime

import numpy as np
import apache_beam as beam
import annoy
from sklearn.random_projection import gaussian_random_matrix

import tensorflow.compat.v1 as tf
import tensorflow_transform as tft
import tensorflow_hub as hub
import tensorflow_transform.beam as tft_beam
print('TF version: {}'.format(tf.__version__))
print('TF-Hub version: {}'.format(hub.__version__))
print('TF-Transform version: {}'.format(tft.__version__))
print('Apache Beam version: {}'.format(beam.__version__))
TF version: 2.1.0
TF-Hub version: 0.8.0
TF-Transform version: 0.21.2
Apache Beam version: 2.19.0

1. Download Sample Data

A Million News Headlines dataset contains news headlines published over a period of 15 years sourced from the reputable Australian Broadcasting Corp. (ABC). This news dataset has a summarised historical record of noteworthy events in the globe from early-2003 to end-2017 with a more granular focus on Australia.

Format: Tab-separated two-column data: 1) publication date and 2) headline text. We are only interested in the headline text.

wget 'https://dataverse.harvard.edu/api/access/datafile/3450625?format=tab&gbrecs=true' -O raw.tsv
wc -l raw.tsv
head raw.tsv
--2020-04-10 14:10:53--  https://dataverse.harvard.edu/api/access/datafile/3450625?format=tab&gbrecs=true
Resolving dataverse.harvard.edu (dataverse.harvard.edu)... 206.191.184.198
Connecting to dataverse.harvard.edu (dataverse.harvard.edu)|206.191.184.198|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 57600231 (55M) [text/tab-separated-values]
Saving to: ‘raw.tsv’

raw.tsv             100%[===================>]  54.93M  13.7MB/s    in 5.0s    

2020-04-10 14:10:59 (10.9 MB/s) - ‘raw.tsv’ saved [57600231/57600231]

1103664 raw.tsv
publish_date    headline_text
20030219    "aba decides against community broadcasting licence"
20030219    "act fire witnesses must be aware of defamation"
20030219    "a g calls for infrastructure protection summit"
20030219    "air nz staff in aust strike for pay rise"
20030219    "air nz strike to affect australian travellers"
20030219    "ambitious olsson wins triple jump"
20030219    "antic delighted with record breaking barca"
20030219    "aussie qualifier stosur wastes four memphis match"
20030219    "aust addresses un security council over iraq"

For simplicity, we only keep the headline text and remove the publication date

!rm -r corpus
!mkdir corpus

with open('corpus/text.txt', 'w') as out_file:
  with open('raw.tsv', 'r') as in_file:
    for line in in_file:
      headline = line.split('\t')[1].strip().strip('"')
      out_file.write(headline+"\n")
rm: cannot remove 'corpus': No such file or directory

tail corpus/text.txt
severe storms forecast for nye in south east queensland
snake catcher pleads for people not to kill reptiles
south australia prepares for party to welcome new year
strikers cool off the heat with big win in adelaide
stunning images from the sydney to hobart yacht
the ashes smiths warners near miss liven up boxing day test
timelapse: brisbanes new year fireworks
what 2017 meant to the kids of australia
what the papodopoulos meeting may mean for ausus
who is george papadopoulos the former trump campaign aide

Helper function to load a TF-Hub module

def load_module(module_url):
  embed_module = hub.Module(module_url)
  placeholder = tf.placeholder(dtype=tf.string)
  embed = embed_module(placeholder)
  session = tf.Session()
  session.run([tf.global_variables_initializer(), tf.tables_initializer()])
  print('TF-Hub module is loaded.')

  def _embeddings_fn(sentences):
    computed_embeddings = session.run(
        embed, feed_dict={placeholder: sentences})
    return computed_embeddings

  return _embeddings_fn

2. Generate Embeddings for the Data.

In this tutorial, we use the Universal Sentence Encoder to generate emebeddings for the headline data. The sentence embeddings can then be easily used to compute sentence level meaning similarity. We run the embedding generation process using Apache Beam and TF-Transform.

Embedding extraction method

encoder = None

def embed_text(text, module_url, random_projection_matrix):
  # Beam will run this function in different processes that need to
  # import hub and load embed_fn (if not previously loaded)
  global encoder
  if not encoder:
    encoder = hub.Module(module_url)
  embedding = encoder(text)
  if random_projection_matrix is not None:
    # Perform random projection for the embedding
    embedding = tf.matmul(
        embedding, tf.cast(random_projection_matrix, embedding.dtype))
  return embedding

Make TFT preprocess_fn method

def make_preprocess_fn(module_url, random_projection_matrix=None):
  '''Makes a tft preprocess_fn'''

  def _preprocess_fn(input_features):
    '''tft preprocess_fn'''
    text = input_features['text']
    # Generate the embedding for the input text
    embedding = embed_text(text, module_url, random_projection_matrix)
    
    output_features = {
        'text': text, 
        'embedding': embedding
        }
        
    return output_features
  
  return _preprocess_fn

Create dataset metadata

def create_metadata():
  '''Creates metadata for the raw data'''
  from tensorflow_transform.tf_metadata import dataset_metadata
  from tensorflow_transform.tf_metadata import schema_utils
  feature_spec = {'text': tf.FixedLenFeature([], dtype=tf.string)}
  schema = schema_utils.schema_from_feature_spec(feature_spec)
  metadata = dataset_metadata.DatasetMetadata(schema)
  return metadata

Beam pipeline

def run_hub2emb(args):
  '''Runs the embedding generation pipeline'''

  options = beam.options.pipeline_options.PipelineOptions(**args)
  args = namedtuple("options", args.keys())(*args.values())

  raw_metadata = create_metadata()
  converter = tft.coders.CsvCoder(
      column_names=['text'], schema=raw_metadata.schema)

  with beam.Pipeline(args.runner, options=options) as pipeline:
    with tft_beam.Context(args.temporary_dir):
      # Read the sentences from the input file
      sentences = ( 
          pipeline
          | 'Read sentences from files' >> beam.io.ReadFromText(
              file_pattern=args.data_dir)
          | 'Convert to dictionary' >> beam.Map(converter.decode)
      )

      sentences_dataset = (sentences, raw_metadata)
      preprocess_fn = make_preprocess_fn(args.module_url, args.random_projection_matrix)
      # Generate the embeddings for the sentence using the TF-Hub module
      embeddings_dataset, _ = (
          sentences_dataset
          | 'Extract embeddings' >> tft_beam.AnalyzeAndTransformDataset(preprocess_fn)
      )

      embeddings, transformed_metadata = embeddings_dataset
      # Write the embeddings to TFRecords files
      embeddings | 'Write embeddings to TFRecords' >> beam.io.tfrecordio.WriteToTFRecord(
          file_path_prefix='{}/emb'.format(args.output_dir),
          file_name_suffix='.tfrecords',
          coder=tft.coders.ExampleProtoCoder(transformed_metadata.schema))

Generaring Random Projection Weight Matrix

Random projection is a simple, yet powerfull technique used to reduce the dimensionality of a set of points which lie in Euclidean space. For a theoretical background, see the Johnson-Lindenstrauss lemma.

Reducing the dimensionality of the embeddings with random projection means less time needed to build and query the ANN index.

In this tutorial we use Gaussian Random Projection from the Scikit-learn library.

def generate_random_projection_weights(original_dim, projected_dim):
  random_projection_matrix = None
  if projected_dim and original_dim > projected_dim:
    random_projection_matrix = gaussian_random_matrix(
        n_components=projected_dim, n_features=original_dim).T
    print("A Gaussian random weight matrix was creates with shape of {}".format(random_projection_matrix.shape))
    print('Storing random projection matrix to disk...')
    with open('random_projection_matrix', 'wb') as handle:
      pickle.dump(random_projection_matrix, 
                  handle, protocol=pickle.HIGHEST_PROTOCOL)
        
  return random_projection_matrix

Set parameters

If you want to build an index using the original embedding space without random projection, set the projected_dim parameter to None. Note that this will slow down the indexing step for high-dimensional embeddings.

module_url = 'https://tfhub.dev/google/universal-sentence-encoder/2' 
projected_dim = 64  

Run pipeline

import tempfile

output_dir = pathlib.Path(tempfile.mkdtemp())
temporary_dir = pathlib.Path(tempfile.mkdtemp())

g = tf.Graph()
with g.as_default():
  original_dim = load_module(module_url)(['']).shape[1]
  random_projection_matrix = None

  if projected_dim:
    random_projection_matrix = generate_random_projection_weights(
        original_dim, projected_dim)

args = {
    'job_name': 'hub2emb-{}'.format(datetime.utcnow().strftime('%y%m%d-%H%M%S')),
    'runner': 'DirectRunner',
    'batch_size': 1024,
    'data_dir': 'corpus/*.txt',
    'output_dir': output_dir,
    'temporary_dir': temporary_dir,
    'module_url': module_url,
    'random_projection_matrix': random_projection_matrix,
}

print("Pipeline args are set.")
args
INFO:tensorflow:Saver not created because there are no variables in the graph to restore

INFO:tensorflow:Saver not created because there are no variables in the graph to restore

TF-Hub module is loaded.
A Gaussian random weight matrix was creates with shape of (512, 64)
Storing random projection matrix to disk...
Pipeline args are set.

/home/kbuilder/.local/lib/python3.6/site-packages/sklearn/utils/deprecation.py:87: FutureWarning: Function gaussian_random_matrix is deprecated; gaussian_random_matrix is deprecated in 0.22 and will be removed in version 0.24.
  warnings.warn(msg, category=FutureWarning)

{'job_name': 'hub2emb-200410-141132',
 'runner': 'DirectRunner',
 'batch_size': 1024,
 'data_dir': 'corpus/*.txt',
 'output_dir': PosixPath('/tmp/tmpp4bkl05s'),
 'temporary_dir': PosixPath('/tmp/tmp1kff0anb'),
 'module_url': 'https://tfhub.dev/google/universal-sentence-encoder/2',
 'random_projection_matrix': array([[-0.1418625 ,  0.07604557,  0.08399526, ...,  0.18551244,
         -0.18382191, -0.03783611],
        [-0.09331975,  0.05518437, -0.02040763, ..., -0.07523481,
          0.10295377,  0.03395856],
        [ 0.10627273, -0.04790061,  0.15580233, ...,  0.02953089,
         -0.01385009, -0.04397705],
        ...,
        [-0.18883092, -0.33584531, -0.0131968 , ...,  0.01451938,
         -0.09707073,  0.18286603],
        [-0.01561171, -0.12631165,  0.04874103, ...,  0.00132466,
          0.00936905, -0.08694398],
        [-0.14257533,  0.06863545, -0.0177571 , ...,  0.22034482,
         -0.02564502, -0.07739396]])}
!rm -r {output_dir}
!rm -r {temporary_dir}

print("Running pipeline...")
%time run_hub2emb(args)
print("Pipeline is done.")
WARNING:apache_beam.runners.interactive.interactive_environment:Dependencies required for Interactive Beam PCollection visualization are not available, please use: `pip install apache-beam[interactive]` to install necessary dependencies to enable all data visualization features.

Running pipeline...
WARNING:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 

Warning:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 

Warning:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 

Warning:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 

INFO:tensorflow:Saver not created because there are no variables in the graph to restore

INFO:tensorflow:Saver not created because there are no variables in the graph to restore

Warning:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow_core/python/saved_model/signature_def_utils_impl.py:201: build_tensor_info (from tensorflow.python.saved_model.utils_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This function will only be available through the v1 compatibility library as tf.compat.v1.saved_model.utils.build_tensor_info or tf.compat.v1.saved_model.build_tensor_info.

Warning:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow_core/python/saved_model/signature_def_utils_impl.py:201: build_tensor_info (from tensorflow.python.saved_model.utils_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This function will only be available through the v1 compatibility library as tf.compat.v1.saved_model.utils.build_tensor_info or tf.compat.v1.saved_model.build_tensor_info.

INFO:tensorflow:Assets added to graph.

INFO:tensorflow:Assets added to graph.

INFO:tensorflow:No assets to write.

INFO:tensorflow:No assets to write.

INFO:tensorflow:SavedModel written to: /tmp/tmp1kff0anb/tftransform_tmp/10efb4e2879343d9beef23746bd26d20/saved_model.pb

INFO:tensorflow:SavedModel written to: /tmp/tmp1kff0anb/tftransform_tmp/10efb4e2879343d9beef23746bd26d20/saved_model.pb

Warning:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 

Warning:tensorflow:Tensorflow version (2.1.0) found. Note that Tensorflow Transform support for TF 2.0 is currently in beta, and features such as tf.function may not work as intended. 
WARNING:apache_beam.io.tfrecordio:Couldn't find python-snappy so the implementation of _TFRecordUtil._masked_crc32c is not as fast as it could be.

CPU times: user 1min 50s, sys: 9.32 s, total: 2min
Wall time: 1min 42s
Pipeline is done.

ls {output_dir}
emb-00000-of-00001.tfrecords

Read some of the generated embeddings...

import itertools

embed_file = os.path.join(output_dir, 'emb-00000-of-00001.tfrecords')
sample = 5
record_iterator =  tf.io.tf_record_iterator(path=embed_file)
for string_record in itertools.islice(record_iterator, sample):
  example = tf.train.Example()
  example.ParseFromString(string_record)
  text = example.features.feature['text'].bytes_list.value
  embedding = np.array(example.features.feature['embedding'].float_list.value)
  print("Embedding dimensions: {}".format(embedding.shape[0]))
  print("{}: {}".format(text, embedding[:10]))

WARNING:tensorflow:From <ipython-input-18-3d6f4d54c65b>:5: tf_record_iterator (from tensorflow.python.lib.io.tf_record) is deprecated and will be removed in a future version.
Instructions for updating:
Use eager execution and: 
`tf.data.TFRecordDataset(path)`

Warning:tensorflow:From <ipython-input-18-3d6f4d54c65b>:5: tf_record_iterator (from tensorflow.python.lib.io.tf_record) is deprecated and will be removed in a future version.
Instructions for updating:
Use eager execution and: 
`tf.data.TFRecordDataset(path)`

Embedding dimensions: 64
[b'headline_text']: [-0.06381369 -0.18098828  0.06505597  0.02521572 -0.08622124  0.13111451
 -0.0725691   0.03183383  0.35552731  0.05419653]
Embedding dimensions: 64
[b'aba decides against community broadcasting licence']: [-0.07446624  0.16849381  0.01514749  0.02923395  0.0652388  -0.1933727
 -0.10387956  0.02826558 -0.17433971  0.11503377]
Embedding dimensions: 64
[b'act fire witnesses must be aware of defamation']: [-0.07213803 -0.00704198 -0.01556329 -0.15588315  0.05151989 -0.17196529
 -0.00918518 -0.22063001 -0.15647726  0.08077872]
Embedding dimensions: 64
[b'a g calls for infrastructure protection summit']: [ 0.02416231  0.04026171 -0.13804214 -0.06155462  0.10524569 -0.02883469
 -0.08305833  0.10515894  0.02298284 -0.13289261]
Embedding dimensions: 64
[b'air nz staff in aust strike for pay rise']: [ 0.14603204 -0.06195154 -0.04195797 -0.12555845 -0.10067023 -0.16411237
 -0.10835019 -0.03645138 -0.1163589   0.06587349]

3. Build the ANN Index for the Embeddings

ANNOY (Approximate Nearest Neighbors Oh Yeah) is a C++ library with Python bindings to search for points in space that are close to a given query point. It also creates large read-only file-based data structures that are mmapped into memory. It is built and used by Spotify for music recommendations.

def build_index(embedding_files_pattern, index_filename, vector_length, 
    metric='angular', num_trees=100):
  '''Builds an ANNOY index'''

  annoy_index = annoy.AnnoyIndex(vector_length, metric=metric)
  # Mapping between the item and its identifier in the index
  mapping = {}

  embed_files = tf.gfile.Glob(embedding_files_pattern)
  print('Found {} embedding file(s).'.format(len(embed_files)))

  item_counter = 0
  for f, embed_file in enumerate(embed_files):
    print('Loading embeddings in file {} of {}...'.format(
      f+1, len(embed_files)))
    record_iterator = tf.io.tf_record_iterator(
      path=embed_file)

    for string_record in record_iterator:
      example = tf.train.Example()
      example.ParseFromString(string_record)
      text = example.features.feature['text'].bytes_list.value[0].decode("utf-8")
      mapping[item_counter] = text
      embedding = np.array(
        example.features.feature['embedding'].float_list.value)
      annoy_index.add_item(item_counter, embedding)
      item_counter += 1
      if item_counter % 100000 == 0:
        print('{} items loaded to the index'.format(item_counter))

  print('A total of {} items added to the index'.format(item_counter))

  print('Building the index with {} trees...'.format(num_trees))
  annoy_index.build(n_trees=num_trees)
  print('Index is successfully built.')
  
  print('Saving index to disk...')
  annoy_index.save(index_filename)
  print('Index is saved to disk.')
  print("Index file size: {} GB".format(
    round(os.path.getsize(index_filename) / float(1024 ** 3), 2)))
  annoy_index.unload()

  print('Saving mapping to disk...')
  with open(index_filename + '.mapping', 'wb') as handle:
    pickle.dump(mapping, handle, protocol=pickle.HIGHEST_PROTOCOL)
  print('Mapping is saved to disk.')
  print("Mapping file size: {} MB".format(
    round(os.path.getsize(index_filename + '.mapping') / float(1024 ** 2), 2)))
embedding_files = "{}/emb-*.tfrecords".format(output_dir)
embedding_dimension = projected_dim
index_filename = "index"

!rm {index_filename}
!rm {index_filename}.mapping

%time build_index(embedding_files, index_filename, embedding_dimension)
rm: cannot remove 'index': No such file or directory
rm: cannot remove 'index.mapping': No such file or directory
Found 1 embedding file(s).
Loading embeddings in file 1 of 1...
100000 items loaded to the index
200000 items loaded to the index
300000 items loaded to the index
400000 items loaded to the index
500000 items loaded to the index
600000 items loaded to the index
700000 items loaded to the index
800000 items loaded to the index
900000 items loaded to the index
1000000 items loaded to the index
1100000 items loaded to the index
A total of 1103664 items added to the index
Building the index with 100 trees...
Index is successfully built.
Saving index to disk...
Index is saved to disk.
Index file size: 1.69 GB
Saving mapping to disk...
Mapping is saved to disk.
Mapping file size: 50.61 MB
CPU times: user 4min 26s, sys: 2.79 s, total: 4min 29s
Wall time: 4min 28s

ls
corpus  index.mapping         raw.tsv
index   random_projection_matrix  semantic_approximate_nearest_neighbors.ipynb

4. Use the Index for Similarity Matching

Now we can use the ANN index to find news headlines that are semantically close to an input query.

Load the index and the mapping files

index = annoy.AnnoyIndex(embedding_dimension)
index.load(index_filename, prefault=True)
print('Annoy index is loaded.')
with open(index_filename + '.mapping', 'rb') as handle:
  mapping = pickle.load(handle)
print('Mapping file is loaded.')

Annoy index is loaded.

/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/ipykernel_launcher.py:1: FutureWarning: The default argument for metric will be removed in future version of Annoy. Please pass metric='angular' explicitly.
  """Entry point for launching an IPython kernel.

Mapping file is loaded.

Similarity matching method

def find_similar_items(embedding, num_matches=5):
  '''Finds similar items to a given embedding in the ANN index'''
  ids = index.get_nns_by_vector(
  embedding, num_matches, search_k=-1, include_distances=False)
  items = [mapping[i] for i in ids]
  return items

Extract embedding from a given query

# Load the TF-Hub module
print("Loading the TF-Hub module...")
g = tf.Graph()
with g.as_default():
  embed_fn = load_module(module_url)
print("TF-Hub module is loaded.")

random_projection_matrix = None
if os.path.exists('random_projection_matrix'):
  print("Loading random projection matrix...")
  with open('random_projection_matrix', 'rb') as handle:
    random_projection_matrix = pickle.load(handle)
  print('random projection matrix is loaded.')

def extract_embeddings(query):
  '''Generates the embedding for the query'''
  query_embedding =  embed_fn([query])[0]
  if random_projection_matrix is not None:
    query_embedding = query_embedding.dot(random_projection_matrix)
  return query_embedding
Loading the TF-Hub module...
INFO:tensorflow:Saver not created because there are no variables in the graph to restore

INFO:tensorflow:Saver not created because there are no variables in the graph to restore

TF-Hub module is loaded.
TF-Hub module is loaded.
Loading random projection matrix...
random projection matrix is loaded.

extract_embeddings("Hello Machine Learning!")[:10]
array([ 0.02640045, -0.09456791,  0.01126807, -0.21123106, -0.00195626,
        0.02094386,  0.0837211 , -0.07892486, -0.08466703,  0.09215297])

Enter a query to find the most similar items


query = "confronting global challenges" 
print("Generating embedding for the query...")
%time query_embedding = extract_embeddings(query)

print("")
print("Finding relevant items in the index...")
%time items = find_similar_items(query_embedding, 10)

print("")
print("Results:")
print("=========")
for item in items:
  print(item)
Generating embedding for the query...
CPU times: user 30.5 ms, sys: 27.8 ms, total: 58.2 ms
Wall time: 7.34 ms

Finding relevant items in the index...
CPU times: user 6.53 ms, sys: 392 µs, total: 6.93 ms
Wall time: 876 µs

Results:
=========
confronting global challenges
bluescope ponders global challenges
nff challenges social media interpretation of
downer challenges un to follow aust example
challenges to austs future
research challenges ice crisis belief
xenophobia exposed
remote education conference to tackle skill
mock disaster tests emergency response
summit to highlight volunteering issues

Want to learn more?

You can learn more about TensorFlow at tensorflow.org and see the TF-Hub API documentation at tensorflow.org/hub. Find available TensorFlow Hub modules at tfhub.dev including more text embedding modules and image feature vector modules.

Also check out the Machine Learning Crash Course which is Google's fast-paced, practical introduction to machine learning.