I think you'll need to clarify what you mean by "recommendations". Do you plan to use your methods for making a recommender system or you want to list up pros and cons of both methods and recommend them for some other tasks? You also ask for classification methods which makes the question even more ambiguous. Please give more details about what you have done and what do you want to do next.

For classification, you can use standard evaluation metrics like F-score and accuracy.

If its a probabilistic model, you can go for Pearson Correlation Score.

may be interseting for you

Chapter 1. Mining Twitter: Exploring Trending Topics, Discovering What People Are Talking About, and More

http://chimera.labs.oreilly.com/books/1234000001583/ch01.html

I would also ask the same questions as Rafal. I've worked with opinion mining and would interpret a positive recommendation as a positive emotion of a tweet. And negative emotion as a negative emotion. SVM or NaiveBayes are possible algorithms for classification of texts.

Possibly we can use c4.5 with prior preprocessing

I used LingPipe for sentiment analysis and it worked pretty well. It is open source and well documented. It has implementations of all the popular NLP algorithms. For me Logistic Regression worked best, Bayesian models and multilayer perceptron had the lowest accuracy. May be this is of interest:

Article Opinion Mining in Social Big Data

I too agree with peter. Even i used some time LingPipe for opinion mining

I have worked with opinion mining and would interpret with strongly positive,positive,neutral,strongly negative and negative as five likert emotion of tweet. Naïve Bayes (NB) Classifier and Maximum Entropy (MaxEnt) Classifier along with optimization iteration algorithms i.e., Generalized Iterative Scaling (GIS) and Improved Iterative Scaling (IIS) can be use.

A Machine Learning Based Approach for Opinion Mining on Social Network Data -

https://link.springer.com/content/pdf/10.1007/978-981-10-3226-4_13.pdf

First let me correct your question classification and recommendation are two different things. Classification can be used for recommendation. so you mean classification algorithms. So there are too many classification algorithm or techniques that can be used for text. Naive Bayes classifier is suitable for text and you can use Multinomial Naive Bayes for tweets with a good accuracy.