Do people like our company, our products, our campaigns, our service, or do they dislike us? This is a question that is crucial for marketing, because who is going to become a customer of a company they don’t like? Sentiment analysis was created to answer the question of favor or disfavor. Here, we’ve compiled for you what exactly it is, how it works, and how you and your company can benefit from it.
The word “sentiment” comes from the French and simply means a feeling or perception. Sentiment analysis investigates which feelings prevail about a certain thing.
Those who object to the term sentiment analysis can use tonality analysis, much in the sense of “setting the tone.” How is a certain thing spoken about? Favorably or unfavorably?
Sentiment analysis is present in text mining and on the stock market.
Some stock market gurus don’t only examine stock charts and economic data, but also the mood of investors. In doing so they want to draw conclusions about how the market will evolve.
But for marketing, sentiment analysis in the context of text mining is vital. For that reason, we will disregard the stock market for this post.
When the relevant core information is filtered out of texts using statistic and linguistic methods, that is called text mining.
In marketing, sentiment analysis predominantly falls under the realm of social media monitoring. Here the term “social media” doesn’t refer only to Facebook, Twitter and the others, but also to YouTube, product and service reviews in shops and portals, as well as forum posts. Today, opinions are very quickly formed and spread through these channels. So those who know how their own products are being talked about in comments and posts can react accordingly. Sentiment analysis isn’t limited to just texts though. Videos, images and even podcasts are closely examined as well.
Sentiment analysis ascertains the mood on social media pertaining to products, service activities, campaigns and companies. Where opinions are predominantly negative, the company can analyze the reasons and react.
A service hotline is being badmouthed. The cause of this needs to be discovered. Is it due to a lack of service staff friendliness or competence? Are the waiting times to long? Is only insufficient assistance being offered? The cause analysis shows that the wait times are too long. To remedy this, more service personnel can be assigned to the project. In addition, other methods for quickly getting in touch, like for example a contact form or chats, can be offered and/or more prominently advertised. Afterwards, further sentiment analysis gives an indication of whether these measures have worked. If that is the case, satisfaction with the hotline will increase.
Sentiment analysis is especially important for marketing campaigns which predominantly focus on social networks. Are the viral videos being liked? Is the podcast well-received? Does the community like the Facebook posts? If the mood is negative, the campaign can be corrected.
In general, we differentiate between manual and automated analysis.
Here, people undertake the inspection of the relevant data. They rate the opinions expressed on social media in regards to positive, negative or neutral tonality.
With automated analysis, software searches through all available data and performs a classification. In doing so, mainly procedures based on linguistic sources are used or the concept of machine learning is applied.
Here, the data is assessed on the basis of pre-determined positive and negative signal words.
The employees were friendly and competent, but the wait time was awful.
The words “friendly” and competent” are evaluated as positive signal words, and the word “awful” as negative. Overall, the sentence is assessed as positive by the software, because two positive signal words are opposed by only one negative.
Problems with this method are the ambiguity of many words, slang terms, and that it doesn’t understand irony and context. For example a long battery service life is something positive, but assigning “long” a positive signal word status is difficult. In the context of wait time, it is of course negative.
Here, software is taught via sample data whether a positive or negative statement is indicated. Once it has this knowledge, it can analyze unfamiliar text and determine its tonality. The sample data is critical with this method. If the software was trained with the aid of customer reviews about cars, then it will fail when evaluating film critiques.
Naturally, even the best software can’t understand all the nuances of human speech and classify the tonality correctly. For this people are still necessary.
Sentiment analysis can be applied in the following areas:
The Miller car dealership is running a campaign on Facebook, and wants to know how well it is being received by its followers. As a general rule, it is sufficient when a Miller dealership employee examines the comments on the social network, and compiles the positive and negative opinions in a spreadsheet.
This procedure of course no longer works with large campaigns or data sets. Here, social monitoring tools are usually employed, or the job is delegated to external service providers like clickworker.
Because clickworker can manually examine even large amounts of data, thanks to their many crowd workers (so-called clickworkers), high-quality results can be achieved in a short amount of time. Combined approaches with data pre-sorted by computer and manually refined are of course also possible.
Thanks to sentiment analysis, you can quickly find out how well marketing campaigns, services, your company or its products are received by people. You don’t need to conduct costly and often not very popular opinion polls. You simply need to analyze the data available on the internet. Artificial intelligence or service providers like clickworker with their enormous manpower can help with that. Those who know their users’ opinions have the opportunity to quickly detect negative developments, and take the appropriate corrective action.
No related posts.
Dieser Artikel wurde am 14.March 2017 von Thomas geschrieben.