11'2 FeatureBased Opinion Mining and Summarization

1
11.2 Feature-Based Opinion Mining and
Summarization
2
Feature-Based Opinion Mining and Summarization

• The reviewer usually writes both positive and
  negative aspects of the product.
• To obtain detailed aspects
• Identifying and extracting product features.
• the picture quality of this camera is amazing,
• Determining whether the opinions on the features
  are positive, negative or neutral.

3
11.2.1 Problem Definition

• Definition (object)
• An object O is an entity which can be a product,
  person, event, organization, or topic.
• O (T, A), T is a hierarchy or taxonomy of
  components, and A is a set of attributes of O.
• Example 2
• digital camera -gt lens, battery, view-finder,
  etc
• digital camera picture quality, size, weight
• Battery battery life, battery size, battery
  weight

4
Problem Definition

• An object is represented as a tree.
• root is the object itself.
• Each non-root node is a component or
  subcomponent.
• Each link represents a part-of relationship.
• Each node is also associated with a set of
  attributes.

5
Problem Definition

• An opinion can be expressed on any node and any
  attribute of the node.
• Example 3
• I do not like this camera
• the picture quality of this camera is poor
• the battery of this camera is bad
• the battery life of this camera is too short.

6
Problem Definition

• To simplify our discussion, we use the word
  features to represent both components and
  attributes.
• Let the evaluative text (e.g., a product review)
  be r.
• r consists of a sequence of sentences.

7
Problem Definition

• Definition (explicit and implicit feature)
• If a feature f appears in evaluative text r, it
  is called an explicit feature in r.
• Example 4
• The battery life of this camera is too short.
  (battery life)
• This camera is too large (size)

8
Problem Definition

• Definition (opinion passage on a feature)
• The opinion passage on feature f of an object
  evaluated in r is a group of consecutive
  sentences that expresses a positive or negative
  opinion on f.
• The picture quality is good, but the battery
  life is short.

9
Problem Definition

• Definition (explicit and implicit opinion)
• explicit opinion a subjective sentence.
• The picture quality of this camera is amazing.
• implicit opinion an objective sentence.
• The earphone broke in two days.
• Definition (opinion holder)
• The holder of a particular opinion is a person or
  an organization that holds the opinion.

10
Problem Definition

• To define a model of an object and a set of
  opinions on the object.
• An object is represented with a finite set of
  features,
• Each feature fi in F can be expressed with a
  finite set of words or phrases Wi, which are
  synonyms.
• Each opinion holder j comments on a subset of the
  features
• For each feature that j comments on, he
  chooses a word or phrase from Wk to describe the
  feature, and then expresses a opinion on it.

11
Problem Definition

• This model introduces three main practical
  problems.
• Problem 1 F and W are unknown. Then, in opinion
  mining, we need to perform three tasks
• Task 1 Identifying and extracting object
  features that have been commented on in each
  evaluative text
• Task 2 Determining whether the opinions are
  positive , negative or neutral.
• Task 3 Grouping synonyms of features, as
  different people may use different words or
  phrases to express the same feature.

12
Problem Definition

• Problem 2 F is known but W is unknown.
• Task 3 becomes the problem of matching discovered
  features with the set of given features F.
• Problem 3 W is known (then F is also known).
• Only need to perform Task 2.
• Example 6
• A cellular phone company wants to mine customer
  reviews on a few models of its phones. There is
  no need to perform Tasks 1 and 3.

13
Problem Definition

• Output The final output is a set of pairs. Each
  pair is denoted by (f, SO), where f is a feature
  and SO is the semantic orientation.
• To use the results, a simple way is to produce a
• feature-based summary of opinions on the object.

14
Problem Definition

• Example 7 Assume we summarize the reviews of a
  particular digital camera, digital_camera_1.

15
Problem Definition

• The summary can also be visualized using a bar
  chart.

16
Problem Definition
17
Problem Definition

• Four other important issues
• 1. Separation of Opinions on the Object itself
  and its Features.
• 2. Granularity of Analysis.
• At level 1identify opinions on the object itself
  and its attributes.
• At level 2 identify opinions on the major
  components and also opinions on the attributes of
  the components.
• At other levels, similar tasks can be performed.

18
Problem Definition

• Example 8
• I like this camera. Its picture quality is
  amazing. However, the battery life is a little
  short
• 3. Opinion Holder Identification
• Opinion holders are more useful for news
  articles, in which the person or organization
  that expressed an opinion is usually stated in
  the text explicitly.
• The opinion holders are often the authors of
  discussion posts, bloggers, or reviewers, whose
  login ids are often known although their true
  identities in the real-world may be unknown.

19
Problem Definition

• 4. Opinioned Object Identification and Pronoun
  Resolution.
• I have a Canon S50 camera purchased from Amazon.
  It takes great photos.
• (1) what object does the post praise?
• (2) what it means in the second sentence?
• to automatically discover answers is a very
  challenging problem.

20
11.2.2 Object Feature Extraction

• There are three main review formats on the Web.
• Different review formats may need different
  techniques to perform the feature extraction
  task.
• Format 1 - Pros, cons and the detailed review

21
Object Feature Extraction

• Format 2 - Pros and cons

22
Object Feature Extraction

• Format 3 - Free format

23
Object Feature Extraction

• In both formats 1 and 2, only product features
  need to be identified.
• For format 3, we need to identify both product
  features and opinion orientations.

24
11.2.3 Feature Extraction from Pros and Cons of
Format 1

• A product feature can be expressed with a noun,
  adjective, verb or adverb.
• The labels and their POS tags used in mining LSRs
  are feature, NN, feature, JJ, feature,
  VB and feature, RB
• Each sentence segment in pros and cons contains
  only one feature. Sentence segments are separated
  by commas, periods, semi-colons, hyphens, s,
  ands, buts, etc.

25
Feature Extraction from Pros and Cons of Format 1

• We call a word that indicates an implicit feature
  an implicit feature indicator.

26
Feature Extraction from Pros and Cons of Format 1

• 1. Training data preparation for LSR mining
• Part-Of-Speech (POS) tagging and sequence
  generation
• Included memory is stingy -gt
• ltincluded, VBmemory, NNis, VBstingy, JJgt.
• Replace the actual feature words with feature,
  lttaggt
• ltincluded, VBfeature, NNis, VBstingy, JJgt

27
Feature Extraction from Pros and Cons of Format 1

• Use an n-gram to produce shorter segments from
  long ones
• generate two trigram sequences
• ltincluded, VBfeature, NNis, VBgt
• ltfeature, NNis, VBstingy, JJgt
• Perform word stemming.

28
Feature Extraction from Pros and Cons of Format 1

• 2. Label sequential rule mining
• A LSR mining system is applied to find rules.
• A suitable minimum confidence and minimum support
  should be used.
• lteasy, JJ to, VBgt
• ?lteasy, JJtofeature, VBgt(language
  pattern)

29
Feature Extraction from Pros and Cons of Format 1

• 3. Feature extraction
• the word in the sentence segment that matches
  feature in a language pattern is extracted.
• Three situations are considered in extraction
• If a sentence segment satisfies multiple rules,
  we search for a matching rule in the following
  order feature, NN, feature, JJ,
  feature, VB and feature, RB.

30
Feature Extraction from Pros and Cons of Format 1

• For sentence segments that no rules apply, nouns
  or noun phrases produced by a POS tagger are
  extracted as features
• For a sentence segment with only a single word,
  the single words are treated as features.

31
Feature Extraction from Pros and Cons of Format 1

• Mapping to Implicit Features
• There are many types of implicit feature
  indicators, Their exact meaning can be domain
  dependent.
• Grouping Synonyms
• It is common that people use different words or
  phrases to describe the same feature.
• Many synonyms are domain dependent.
• Granularity of Features
• In a practical application, we need to determine
  the right level of analysis.

32
11.2.4 Feature Extraction from Reviews of
Formats 2 and 3

• Complete sentences are more complex and contain a
  large amount of irrelevant information.
• An unsupervised method for finding explicit
  features that are nouns and noun phrases.
• This method requires a large number of reviews,
  and consists of two steps
• 1. most product features are nouns, and those
  nouns that are frequently talked about are
  usually genuine and important features.

33
Feature Extraction from Reviews of Formats 2 and 3

• 2. Finding infrequent features by making use of
  sentiment words. The same opinion word can be
  used to describe different objects.
• The precision of step 1 of the above algorithm
  was improved by Popescu and Etzioni.
• by computing a PMI score between the phrase and
  meronymy discriminators associated with the
  product class.

34
Feature Extraction from Reviews of Formats 2 and 3

• where f is a candidate feature identified in step
  1 and d is a discriminator.
• Web search is used to find the number of hits.
• If the PMI value of a candidate feature is too
  low, it may not be a component of the product
  because f and d do not co-occur frequently.

35
11.2.5 Opinion Orientation Classification

• 1. Using sentiment words and phrases
• Manually find a set of seed positive and negative
  words.
• Grow each of the seed set by iteratively
  searching for their synonyms and antonyms in
  WordNet until convergence.
• Manually inspect the results to remove those
  incorrect words.

36
Opinion Orientation Classification

• using the final lists of positive and negative
  words, phrases, idioms and patterns, each
  sentence that contains product features can be
  classified as follows
• A positive word or phrase is assigned a score of
  1 and a negative word or phrase is assigned a
  score of -1. All the scores are then summed up.
• If the final total is positive, then the sentence
  is positive, otherwise it is negative.

37
Opinion Orientation Classification

• If a negation word is near a sentiment word, the
  opinion is reversed.
• A sentence that contains a but clause
  (subsentence that starts with but, however,
  etc.) indicates a sentiment change for the
  feature in the clause.
• The opinion orientations of many words are domain
  and/or sentence context dependent.