How to generalize comments using NLP

Question

I have list of log comments in CSV file. I want to cluster those log comments using K-Means and after that I want convert each cluster comments into general form. for eg. I have bunch of comments in one cluster which starts from "Reservation number failed......" and I want to convert those comments into particular comment like "Reservation failure".

I can achieve this by giving specific name to each cluster after seeing each cluster. But I don't want like this. I want to create intelligent model which automatically create generalized comments for me.

I would not like to assign name to each cluster. Basically I am done with clustering part. that is, I have lets say 3 clusters as below

• cluster 0 : list of comments like "Reservation number failed......", total comments: 15
• cluster 1 : list of comments like "Request timeout failed due to ......", total comments:9
• cluster 2 : list of comments like "Dinning reservation successfully completed...", total comments: 5

I want to build model that intelligently assign name to each cluster by its contents. for eg .

• cluster 0 will get name as "Reservation failure"
• cluster 1 will get name as "Request timeout failure"
• cluster 2 will get name as "Dinning reservation successful"

if after training more data with some different comments. it should create another cluster and assign the name as per content.

Answer 1

Basically, we want to cluster similar comments and assign a name/entity to it. I suggest you to use Doc2Vec to convert the comments to fixed-sized vectors. Each of your comments will then be a n-dimensional vector. Comments with similar words/phrases will lie in the close proximity of one another.

Now, using K-Means Clustering, we can form clusters of vectors that represent comments having a similar meaning. Once the clusters are formed, assign a name to each of them.

For a given sample ( comment ), the model will first transform the given comment into a vector. The model will then check for a cluster that is nearest to the given sample's vector. The output will be the name of the nearest cluster.