I have customer feedback data about 2-3 products from 100 customers. Number of questions are around 160. I have data in excel format. Header row contains the question and row below contains the responses from customers. Not all questions have been answered by each customer. I have to categorize the feedback as either positive or negative. The analysis needs to be done in R.

This data contains boolean responses, numerical responses and text responses.

I've collected the data for text responses in a separate sheet and trying to do sentiment analysis using sentiment analysis package and the general text mining package in R.

How to do the sentiment analysis of boolean and numerical responses ?

Also, Can I combine the score from all the above three analysis and can categorize the response as positive or negative overall ?

  • 2
    $\begingroup$ This is too broad to be answerable as currently written. If you can write a focused, concrete question, we can help you. You may want to read our materials regarding asking questions on CV. $\endgroup$ – gung - Reinstate Monica Jul 9 '14 at 23:24
  • $\begingroup$ This data contains boolean responses, numerical responses and text responses. So should I do sentiment analysis of all these types of responses separately and then categorize the response as positive or negative. Is it possible to do sentiment analysis of all these responses together. $\endgroup$ – Anky Jul 9 '14 at 23:29
  • 1
    $\begingroup$ Edited the question. Please help me out $\endgroup$ – Anky Jul 10 '14 at 0:35

This might be the wrong answer, but I work daily with such datasets in market research. Standard practice looks like this:

  1. Identify variables which could define differences (product, gender, age etc. are obvious candidates)
  2. Identify variables who are of interest, e.g. overall satisfaction. If the variables have a meaningful scale like a 6 point scale recodethem into top 2 / bottom 2 values - the middle values are often of no intereset.
  3. Produces crosstables with the variables from #1 in the columns, and #2 as rows. The advantage of this method is that missing values etc. are not fatal - they reduce the number of cases in some table, but not in all. Disdvantage is the low statistical power of such tests.

  4. If there are text responses, survey the responses (10%) and generate some sort of key like "satisfied customer", "buttons are to small" etc. pp. - then create 0/1 variables for each key-element. Then some poor bastard has to code each response into the key(s). Boring, but doable.

  5. Advanced analysis like regression / correlation or what else seems apropriate.

  • $\begingroup$ Thanks for the reply Christian. I am going to use Bayesian Networks to analyze the customer satisfaction survey data. Now I am facing one problem. I have a column which has a data set in 5 point scale i.e 3 to 7 with 3 being dissatisfied and 5 being highly satisfied. I have to convert this data to 2 point scale i.e either Yes or No. Any inputs on this issue will be highly appreciated. $\endgroup$ – Anky Jul 11 '14 at 19:03
  • $\begingroup$ 3 to 7 sounds highly unusal. YOu might want to consider using only the extreme 2 values at both ends, e.g. "6/7 -> 1, 3/4 -> 0", the 5 will be discarded, but is often of less interest because neutral answer have often not much sway anyway. An alternative would be to take the route: 6/7 ->1 , else -> 0, but the size of the categories is unequal which I doesn't like - but some people like this more, because no answers are lost. Probaly try both methods and take a look at the results. $\endgroup$ – Christian Sauer Jul 12 '14 at 6:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.