kdd’04 august 22-25,2004,seattle,washington,usa mining and summarizing customer reviews bing liu...

KDD’04 AUGUST 22-25,2004,SEATTLE,WASHINGTON,USA

Mining and Summarizing Customer Reviews

BING LIUMINQING HU

AGENDA

1.INTRODUCTION2.RELATED WORK3.THE PROPOSED TECHNIQUES4.EXPERIMENTAL EVALUATION5.CONCLUSIONS

INTRODUCTION(1)

e-commerce is becoming more and more popular, the number of customer reviews that a product receives grows rapidly

Study the problem of generating feature-based summaries of customer reviews of products sold online

(1)identifying features of the product that customers have expressed their opinions on (called product features) (2)for each feature, identifying review sentences that give positive or negative opinions (3)producing a summary using the discovered information

INTRODUCTION(2)

INTRODUCTION(3)

Our task is different from traditional text summarization in a number of ways.

(1) a summary in our case is structured rather than another (but shorter) free text document as produced by most text summarization systems.

(2) we are only interested in features of the product that customers have opinions on and also whether the opinions are positive or negative.

INTRODUCTION(4)

task is performed in three main steps:

(1) Mining product features that have been commented on by customers. We make use of both data mining and natural language processing techniques to perform this task.

(2)Identifying opinion sentences in each review and deciding whether each opinion sentence is positive or negative, these opinion sentences must contain one or more product features identified above (3) Summarizing the results. This step aggregates the results of previous steps and presents them in the format of Figure 1.

2.RELATED WORK

2.1 Subjective Genre Classification

2.2 Sentiment Classification

2.3 Text Summarization

3.TECHNIQUES

3.1 Part-of-Speech Tagging (POS)

Product features are usually nouns or noun phrases in review sentences.

Each sentence is saved in the review database along with the POS tag information of each word in the sentence.-> NLProcessor

Some pre-processing of words is also performed-remove stopwords-stemming -fuzzy matching

3.2 Frequent Features Identification(1)

an easy and a hard sentence from the reviews of a digital camera:

“The pictures are very clear.” -> explicit picture“While light, it will not easily fit in pockets.” -> implicit size

leave finding implicit features to our future work.

Association mining -> association miner CBA-customer review contains many things that are not directly related to product features. However, when they comment on product features, the words that they use converge.

-frequent itemsets are likely to be product features. Those noun/noun phrases that are infrequent are likely to be non-product features

Association Rule Mining

在美國，一些年輕的父親下班後經常要到超市去買嬰兒尿布，超市也因此發現了一個規律，在購買嬰兒尿布的年輕父親們中，有 30% ～ 40% 的人同時要買一些啤酒。超市隨後調整了貨架的擺放，把尿布和啤酒放在一起，明顯增加了銷售額。同樣的，我們還可以根據關聯規則在商品銷售方面做各種促銷活動。

Association Rule: Basic Concepts: 給定︰

Items set: I={i1,i2,...,im} 商品的集合 The task-relevant data D: 是資料庫交易的集合，每個交易 T則是項目

的集合 A, B 為兩個項目集合，交易 T包含 A if and only if

關聯規則是如下蘊涵式︰其中並且，規則在資料集 D中成立，並且具有

支持度 s和置信度 c

IB I,A ΦBA BA c] [s, BA

TA


法則 X Y的支持度 (support) 定義為項目集的支持度。EX:

法則 X Y的信心水準 (confidence) 是符合條件句與結論句的交易個數佔全體符合條件句的交易個數之比例，亦即

信心水準 =

EX: 25 items set {2} 的支持個數為 5, 支持度為 5/10=0.5 items set {2,5} 的支持個數為 3, 支持度為 3/10=0.3 25 的信心水準為 0.3/0.5=0.6

Customerbuys diaper A

Customerbuys both

Customerbuys beer B

交易編號商品編號

1 2, 5, 7

2 1, 3, 4, 6

3 2, 6, 7

4 2, 4, 5

5 3, 6

6 2, 4, 6

7 1, 4, 5

8 1, 3, 5

9 2, 3, 5

10 1, 3, 5

YX

的支持度的支持度

X

YX


We then run the association rule miner, CBA (Liu, Hsu and Ma 1998), which is based on the Apriori algorithm in.

The Apriori algorithm works in two steps.

first: it finds all frequent itemsets from a set of transactions that satisfy a user-specified minimum support. second: it generates rules from the discovered frequent temsets.For our task, we only need the first step

In our work, we define an itemset as frequent if it appears in more than 1% (minimum support) of the review sentences.

EX: 假設我們設定 minimum support 值為 40% ，且資料庫中有10,000 筆交易記錄，則 {AB} 這個 itemsets 所出現的筆數必須大於等於 4,000 （ 10,00040% ）才算 frequent itemsets.

3.2 Frequent Features Identification(2)

However, not all candidate frequent features generated by association mining are genuine features.

(1)Compactness pruning: This method checks features that contain at least two words, which we call feature phrases, and remove those that are likely to be meaningless. -The association mining algorithm does not consider the position of an item (or word) in a sentence.

(2)Redundancy pruning: focus on removing redundant features that contain single words

-p-support of feature ftr is the number of sentences that ftr appears in as a noun or noun phrase, and these sentences must contain no feature phrase that is a superset of ftr.

EX: life by itself is not a useful feature while battery life is a meaningful feature phrase.

Compactness pruning

Let f be a frequent feature phrase and f contains n words. Assume that a sentence s contains f and the sequence of the words in f that appear in s is: w1, w2, …, wn. If the word distance in s between any two adjacent words (wi and wi+1) in the above sequence is no greater than 3, then we say f is compact in s.

If f occurs in m sentences in the review database, and it is compact in at least 2 of the m sentences, then we call f a compact feature phrase.

EX: “I had searched for a digital camera for 3 months.” “This is the best digital camera on the market” “The camera does not have a digital zoom”

3.3 Opinion Words Extraction(1)

These are words that are primarily used to express subjective opinions.

Definition: opinion sentence If a sentence contains one or more product

features and one or more opinion words, then the sentence is called an opinion sentence.

3.3 Opinion Words Extraction(2)

EX: horrible is the effective opinion of strap in “The strap is horrible and gets in the way of parts of the camera you need access to.”

Effective opinions will be useful when we predict the orientation of opinion sentences.

3.4 Orientation Identification for Opinion(1)

For each opinion word, we need to identify its semantic orientation, which will be used to predict the semantic orientation of each opinion sentence.

WordNet do not include semantic orientation information for each word

In general, adjectives share the same orientation as their synonyms and opposite orientations as their antonyms. We use this idea to predict the orientation of an adjective.


enough seed adjectives with known orientations, we can almost predict the orientations of all the adjective words in the review collection.

Our strategy is to use a set of seed adjectives, which we know their orientations and then grow this set by searching in the WordNet.

3.5 Infrequent Feature Identification(1)

These features can also be interesting to some potential customers and the manufacturer of the product. -> generated for completeness

association mining is unable to identify such infrequent features

“The pictures are absolutely amazing.” “The software that comes with it is amazing.”

the same opinion word amazing yet describing different features: sentence 1 is about the pictures, and sentence 2 is about the software. Since one adjective word can be used to describe different objects, we could use the opinion words to look for features that cannot be found in the frequent feature generation step using association mining.

3.5 Infrequent Feature Identification(2)

use the nearest noun/noun phrase as the noun/noun phrase that the opinion word modifies because that is what happens most of the time

it could also find nouns/noun phrases that are irrelevant to the given product, they account for around 15-20% of the total number

Since we rank features according to their p-supports, those wrong infrequent features will be ranked very low and thus will not affect most of the users.

3.6 Predicting the Orientation of Opinion Sentences(1)

we use the dominant orientation of the opinion words in the sentence to determine the orientation of the sentence

where there is the same number of positive and negative opinion words in the sentence, we predict the orientation using the average orientation of effective opinions or the orientation of the previous opinion sentence (recall that effective opinion is the closest opinion word for a feature in an opinion sentence)

3.6 Predicting the Orientation of Opinion Sentences(2)

3.7 Summary Generation

A count is computed to show how many reviews give positive/negative opinions to the feature.

All features are ranked according to the frequency of their appearances in the reviews. Feature phrases appear before single word features as phrases normally are more interesting to users.

4. EXPERIMENTAL EVALUATION(1)

evaluate FBS from three perspectives: 1. ) The effectiveness of feature extraction. 2. ) The effectiveness of opinion sentence

extraction. 3. ) The accuracy of orientation prediction of

opinion sentences.

using the customer reviews of five electronics products: 2 digital cameras, 1 DVD player, 1 mp3 player, and 1 cellular phone., collected from Amazon.com and C|net.com


Most of these terms are not product features at all

FASTR does not find one-word terms, but only term phrases that consist of two or more words.


people like to describe their “stories” with the product lively, there is no indication of whether the user likes the features or not, our system labels these sentences as opinion sentences because they contain both product features and some opinion adjectives. This decreases precision.

• the average accuracy for the five products is 84%. This shows that our method of using WordNet to predict adjective semantic orientations and orientations of opinion sentences are highly effective.


three main limitations of our system: (1) We have not dealt with opinion sentences that need pronoun resolution . EX:“it is quiet but powerful”. what it represents?

(2) We only used adjectives as indicators of opinion orientations of sentences. However, verbs and nouns can also be used for the purpose EX: “I like the feeling of the camera”. “I highly recommend the camera”.

(3) It is also important to study the strength of opinions. Some opinions are very strong and some are quite mild. Highlighting strong opinions (strongly like or dislike) can be very useful for both individual shoppers and product manufacturers.

kdd’04 august 22-25,2004,seattle,washington,usa mining and summarizing customer reviews bing liu...

Documents