The same classification methods explained in the post related to “Document-level sentiment classification” can be applied to individual sentences. But in this case, it’s not assumed each sentence as an opinion. So, in these cases, it’s also necessary a previous task to classify sentences as subjective or objective. The sentences classified as subjective sentences are then classified as positive or negative opinions, what is called sentence-level sentiment classification.

Given a sentence, two sub-tasks are performed:

• Subjectivity classification: Determine whether is a subjective sentence or an objective sentence.
• Sentence-level sentiment classification: If is subjective, determine whether it expresses a positive or negative opinion.

Both problems are classification problems. Thus, traditional supervised learning methods are again applicable.

However, a limitation continues, and it’s the assumption that a sentence expresses a single opinion from a single opinion holder, fact that it’s only appropriate for simple sentences.

Because of this, it’s interesting to split compound sentences in simple sentences using NLP (Natural Language Processing) techniques. However, this is not always an easy task and entail different issues, as anaphora resolution and subject identification, that complicate the sentiment analysis task.

FEATURE-BASED SENTIMENT ANALYSIS

Although classifying opinionated texts at the document level or at the sentence level is useful in many cases, they do not provide the necessary detail needed for some other applications. A positive opinionated document on a particular object does not mean that the author has positive opinions on all aspects or features of the object. Likewise, a negative opinionated document does not mean that the author dislikes everything.

In a typical opinionated text, the author writes both positive and negative aspects of the object, although the general sentiment on the object may be positive or negative. Document-level and sentence-level classification does not provide such information. To obtain such details, we need to go to the object feature level.

For this, we need two key mining tasks:

• Identify object features that have been commented on.

Current research on object feature extraction is mainly carried out in online product reviews. There are two common review formats on the Web.

• Pros, cons and the detailed review: The reviewer is asked to describe Pros and Cons separately and also write a full review.
• Free format: The reviewer can write freely, no separation Pros and Cons.
• Determine whether the opinions on the features are positive, negative or neutral.

To identify the orientation of expressed opinions on an object feature in a sentence the sentence-level and clause-level sentiment classification methods discussed applicable here.

NOTE: Directly expressing positive or negative opinions on an object and its features is only one form of evaluation. Comparing the object with some other similar objects is another. Comparisons are related to but are also quite different from direct opinions. They have not only different semantic meanings, but also different syntactic forms. For example, a typical direct opinion sentence is “The acceleration of this car is amazing”, while a typical comparison sentence is “The acceleration of your car is much better than the acceleration of John’s Porsche”.

A comparative opinion expresses a relation of similarities or differences between two or more objects, and/or object preferences of the opinion holder based on some of the shared features of the objects. A comparative opinion is usually expressed using the comparative or superlative form of an adjective or adverb, although not always.

The treatment of this kind of phrases is different, and many of the available tools don’t evaluate them. In contrast, WebOpinion makes a special treatment of these phrases to extract all the possible opinionated elements of a text, including the comments associated to them.

DOCUMENT-LEVEL SENTIMENT CLASSIFICATION

The first and easiest approach consists of classifying an opinionated document (e.g., a product review) as a positive or negative opinion. This task is commonly known as the document-level sentiment classification because it considers the whole document as the basic information unit.

In the document-level sentiment classification given a set of opinionated documents D, it determines whether each document d ∈ D expresses a positive or negative opinion (or sentiment) on an object.
Existing research and work on sentiment classification assume that an opinionated document d expresses opinions on a single object and the opinions are from a single opinion holder. These assumptions makes that this technique only could be applied in concrete contexts, because it’s difficult to find documents with only opinions on a single object.

Most existing techniques for document-level sentiment classification are based on supervised learning, although there are also some unsupervised methods.
Classification Based on Supervised Learning (document level)Sentiment classification can obviously be formulated as a supervised learning problem with two class labels (positive and negative), assuming that all the documents are known to be opinionated.
Training and testing data used in existing research are mostly product reviews with a reviewer-assigned rating (usually between 1 and 5 stars). Typically, a review with 4-5 stars is considered a positive review (thumbs-up), and a review with 1-2 stars is considered a negative review (thumbs-down).
Existing supervised learning methods can be readily applied to sentiment classification, for example, methods as Naïve Bayes, SVM (Support Vector Machines) or Maximum Entropy have been used with “good” results.
Some basic experiments using this approach, start using unigrams as features in classification with methods as Naïve Bayes or SVM. With the training set a classifier is obtained that using the unigrams that typically appears in positive and negative documents, is able to assign a polarity to a document. With this, looking if the unigrams of a document are closer to the unigrams that typically define positive or negative documents, a polarity will be assigned.
This approach has different limitations, some of them are related to the fact that it cannot take in account negations. This is because, for example, the term “not” will appear in positive and negative opinions, and it won’t be a relevant term to identify polarity. With this, “not good” and “good” will be classified as positive opinions, because “not” won’t be a polar term for the classifier.

To solve this limitation, some experiments pre-process the documents and introduce the negation in a unigram. Changing “not good” by “not good” and training the classifier with this compound term. Other experiments use bigrams and trigrams to consider expressions with more than one word.

Bo Pang took this approach to classify movie reviews into two classes, positive and negative (neutral reviews were not used in this work), having results with precisions close to the 83% using SVM. In his experiments he used as features: unigrams, bigramas, POS tags, tags for the negation and the position of the terms.
Other authors and experiments, during the pre-processing step, uses stop-words list to remove them from the documents, and use lemmatization and stemming.
In all the cases, the noise introduced by the part of the documents that is no related to the object that is been evaluated, doesn’t generate very accurate results. An easy solution for this is to use windows of action, using the classifying method only with the text that it’s around the keywords that identify the object that is been evaluated.
One of the bottlenecks and biggest problems in applying supervised learning is the manual effort involved in annotating a large number of training examples.

Classification Based on Unsupervised Learning
Using unsupervised learning based on opinion words and phrases seems to be a quite interesting approach. Was used by authors as P. Turney in “Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews”. The algorithm consists of three steps:

• Step 1: It extracts phrases containing adjectives or adverbs (research has shown that adjectives and adverbs are good indicators of subjectivity and opinions)

• Step 2: It estimates the orientation of the extracted phrases using the Pointwise Mutual Information (PMI)

The opinion orientation (OO) of a phrase is computed based on the sum of the PMI of the different segments. For example:
OO (phrase) = PMI (phrase, “excelent”) – PMI (phrase, “poor”)

• Step 3: Given a review, the final algorithm computes the opinion orientation average of all phrases in the review, and classifies the review as recommended if the average is positive, not recommended otherwise.

Apart from this method many other unsupervised methods exist, but they are not the most widely used methods.
In our next post, we will comment different approaches for sentence-level sentiment classification.

At that time I realized the potential of the product that was in the Online Reputation Management, taking into account the possibility of streamlining the processes and resources and to improve the manual methodology for obtaining information.

WebOpinion is the clearest example of evolutionary improvement of the ORM (Online Reputation Management). With WebOpinion, Online Reputation Management is already a fact, and possible with a single click, with no effort and completely automatic, to know what customers, suppliers and partners are saying about a Company, Brand, Product or Service and even People. It works by tracing the main sources of online information, tracing Blogs, Social Networks, Web 2.0 and Forums.

The “economics” or WebOpinion benefits are many, beginning with an operational cost savings together with improved accuracy and scope of the search results. Additionally, the tool calculates monthly IReOn (Online Reputation Index) automatically, which allows you to monitor monthly how the company / product / brand / person is in terms of online reputation. It is calculated by an algorithm that combines how well or badly clients speak of the company / brand / product / individual, whether the text is long or short with the importance of who made the comments (if we are talking about someone with authority and visibility in the network or not).

One of the clear advantages of WebOpinion is that it significantly shortens the reaction time from those involved in a crisis or disadvantage. Additionally it identifies which users are most active and what is the dissemination and visibility of their comments.

With WebOpinion, Companies can be more calm and can quickly identify and see the views that people put into the network about their brands, products, people, and even themselves as a whole. Once detected, an internal management system is needed, or some kind of support from the media agencies, an important pillar in managing the corrective action network.

Today, at the present time, at this very moment, people may be talking about you in the network, damaging your prestige and reputation. Or on the contrary people may be talking well about you to other potential customers and recommending the virtues of your products. In the measurement of reputation, the well known phrase “It is not important if they speak badly about you or if they speak well about you, but that they speak about you” is not true today. The Reputation Management must seek to achieve excellence with the maximum number of posts with positive feedback and minimizing the negative comments that users post.

For this reason, a company must take care of its reputation as if it were their own subsistence. And beyond the corporate reputation a literal effort is crucial, to get a business leadership, in several fields including corporate social responsibility, quality of its products, responsible advertising, the truth of the communication process, openness of the people representing the Company, employee policies, its valuation in the online channel … it all adds up in the perspective of the client and his clinical eye.

It is therefore necessary but not sufficient to keep customers satisfied, loyal customers and manage their value through personalized and differentiated treatment. Customers cannot be cheated. Clients choose, test, consume, judge and communicate. And they are able to challenge the organizations when they don´t feel fully convinced. Reputation gets convinced and the lack of it “wins without convincing.”

By the year 2010, children join the world with Internet under one arm and the other with a mobile. Web 2.0 has been established in our lives and I’m in love with technology. Social networks, wikis, blogs, forums, applications for mobile perhaps unthinkable a few years ago. All technical machines have been created and launched so that our voice can reach anywhere in the world and anyone can read us, internalize and judge our words. And with that, we have become judges of Companies, and we can rent a special megaphone paying 25 euros per month per 3 Megabites guaranteed and 20 euros per mobile plan. Technology has provided an excellent opportunity as clients to share their life experiences on the network and has given us an excellent opportunity as entrepreneurs to manage that business reputation through specific tools, to help us quickly identify sources of problems.

Talking about tools and taking into account the ease of online opinion on Companies, Brands, Products and People, how can we manage the online communication processes of thousands of consumers and users, who can give their opinion on various sources and express themselves in an informal and in a free way? The answer to this question is called WEBOPINION and it’s the most reliable online reputation management tool that I will be describing in the next post.

1. the opinion that people in general have about someone or something

2. a place in public esteem or regard : good name

Since the beginning of the year 2000, when we still didn’t have the thread of the crisis over our heads, I explained in my seminars to the marketing and business managers in Spanish companies, the different meanings of ‘reputation’, a word that I consider like the “word of the XXI century”.

In the middle nineties, when Internet was no more than 56k and a Netscape Navigator and the cell phones weighted nearly like the actual Mac Book Pro, the Services Companies started having interest in the achievement of the customer satisfaction and in the improvements of the quality services. It’s a challenge thinking that the gas operator comes to your house and he is going to take careful of your Persian carpet by wearing plastic covers on his boots!!

The challenge is done! The companies focused on the customer satisfaction and they tried to identify which the main satisfaction’s reasons were and above all, the dissatisfaction. The companies tried also to evaluate what kind of impact they had on their “happiness” as customers of this company.  All these developed through quantitative, like factorial analysis and multiple regression and of course, qualitative techniques, like the customers Focus Groups.

So, in 2000, the companies mastered the control of the customer’s satisfaction and they started to work on the building of customer loyalty to keep a loyal customer’s bases; that was increased through the called “Word of the Mouth” – or “mouth to ear” like I prefer to call it. The satisfied customer was loyal, because the programmes about the building of customer loyalty were based on “Gemini effect”, that owns an important emotional, rational and economic component. Normally, the customer could win some curious objects exchanging the accumulated points. A satisfied customer is a loyal customer that communicates his happiness to others potential customers…finally, the clients think that a company has the necessary reputation to be still his supplier.

In 2005 the communication by Internet or by mobile phone was more common in our existences. The “mouth to ear” wasn’t only physic, but also online, and so, the unsatisfied customers or the disloyal customers had an important ‘megaphone’ to show their disagreements. And if you want to change to other company was so difficult; above all in the telecommunication mobile sector or Internet business…does it ring you a bell?

The companies’ master motion was to work in the next level of the Business Management, the Customer Value Management, that is, to offer only to the customers with a high value (financial) for the companies the biggest care. It would be necessary a different treatment, now they managed the value to keep and built customer loyalty only of those profitable customers. The management of the resources is very important and the innovation in the processes too.

At the end, the customers value the esteem that they hold the Companies, Products, Brands and Persons, in order of how many the Companies satisfy them, built their loyalty and generate value for them. And if they fail to live up to their expectations and they feel that they are been cheating, then they show its dissatisfaction with comments and opinions that can affect in an important way the reputation of this Company, Product, Brand or Person.

An enormous effort has been developed in opinion mining and the sentiment analysis to get this objective. Both of them concentrate all the efforts in the opinion computational treatment, the sentiment analysis and in the texts. Thanks to the growing interest in all this tools, there has been an authentic explosion in the number of companies that nowadays offer this sentiment analysis services. Just in the USA there are today more than 30 companies that nowadays are working on this area.

In the last years the Web 2.0 explosion had made that a big amount of information is shared through blogs, social networks and collaboration sites.

Information can be sorted in two general groups: facts and opinions.  The facts are expressions and information about companies, events and their properties. In other hand, the opinions are subjective sentiment, sensation and evaluations about companies, events and properties.

Opinions are valuable information for a particular user and for a marketing department in a company. If an online user wants to carry on a purchase it’s useful if the user can check before what the other’s opinion users are, and what idea they have about the brand, the product or service.

It’s also interesting for the company because it’s fundamental to know what the users think about and to reveal which the powers and debilities of your products are.

Few years before it was not so important the investment on the opinion search because the subjective texts nearly didn’t exist. During this period if someone needed to make a decision just asked friends or relatives. If a company wanted this information, it turned to opinion polls or group analysis. However these methods have changed a lot since the growing and development of the websites where the content has improved considerably.

Nowadays, if someone wants to buy a product, he hasn’t only to ask their friends and familiars because a lot of opinions exist about this product in the Web that shows the evaluation of the product made by different users. For a company, instead of carrying-out surveys or to hire people to obtain the customers’ opinions about their products and those of the competition, the content generated in the Web can give it this information.

To understand the importance of the opinions to make a decision, it is enough to analyse how the traditional marketing has reduced its impact on the customers in favour of the opinions. For example, 76% of consumers declare that they don’t believe in the messages that the companies transmit in their advertisements, while, according to the “Word of Mouth Marketing Association”, 92% consider the familiars’ opinions and those founded in Internet the best way to obtain information about the products.

However, to find adequate sources and to monitor them is not a trivial task owning to the enormous amount of sources in the Web, each one with a lot of opinions and feelings. In some cases, these opinions are hidden in blogs and forums’ posts or in the comments added to a new, a video or a source. For a human lector is complicated and laborious to find the important sources, to extract the phrase related with the opinions, read, summarize and organize them in an useful way. For that, the automation in the discovery of opinions and the synthesis systems are needed. The analysis of feeling and the opinion mining systems arise from this necessity.

The analysis of feelings and the opinions mining areas are located between the natural language processing and the web content mining, that try to determine the aptitude of a communicator who is writing or speaking about a subject. This analysis gives you the possibility to obtain an opinion or an evaluation, its state of mind or the emotional intentions of the communication.

These areas put into the practice in the Web are using to detect opinions and feelings about actions and products. You can read comments in forums, social networks and blogs and, then, process them. The algorithms try to identify the words with positive and negative feelings. However, it is not an easy task and the algorithms, sometimes, don’t get right because there are different problems related with the processing of the language as, for instance, the difficulty of detecting automatically sarcastic remarks.

There are two typical ways of doing the research:

1. We can find public opinions about an object or its characteristics. The object could be a product, person, organization, topic or event.
2. We can find the opinions about a person or organizations related to an object or characteristic. That kind of research is interesting when the person or the organizations that express the opinions are indicated. In this way, we can know what the president of the United States, Barak Obama, thinks about other politician or about a brand.

However we cannot identify the author of each web opinion. We have to check determinate sources where the opinion’s author is included.

Like the traditional web’s research, the opinions’ research has two principles tasks:

1. Recuperate important documents/sentences for an user’s research
2. Arrange the relevant documents/sentences or based on some propriety (author, date, polarization).

Also there are important differences. When we recuperate that information or we research the opinions, we have to do two additional works:

1. We have to find relevant documents or sentences for the topic that interest us.
2. We have to determinate if a document or sentence expresses a negative or positive opinion. That work is only for the sentimental analyse and the opinion’s research is more complicated than the traditional research.

Important companies are working hard to grow up and improve the recent techniques to calculate the sentiment and to create a general and robust system to recuperate opinions that could become lead, as is Google in traditional information retrieval systems. The products as WebOpinion try to accomplish this task, using computational algorithms that can revolutionize the present world  of marketing.