looking for best way to implement a Recommendations Without User Preferences. Content-based recommendation as one of content similarity measurement. Let say we have all of article the user can read from , we need to provide each user with a list of top 5 most similar article that are similar to that article . please help with any simple guide , implementation example, case study, any help is appreciated .
Interesting!
I think keywords play a vital role in this case. For example I am reading paper in query expansion and paper's keywords are : W1, W2,.., Wn. These keywords could be used to suggest other similar papers to me in accordance to the words similarity and bondness between paper I am already reading and those in store.
Though it is simple, I wish it works!
- You can you contents of an article to extract topics/keywords or frequent words.
- Use tfidf vectorizer to create a model (remember to remove stopwords).
- Then use cosine similarity to get similar articles.
import pandas as pd
import nltk
nltk.download('stopwords') # download stopwords
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
#your csv contains articles per row-wise.
df = pd.read_csv("path/to/file.csv")
#tokenize articles from sentences to words.
tokenized = nltk.tokenize.word_tokenize(df['column_name'])
tfidf = TfidfVectorizer(stop_words='english')
# This process also builds a vocabulary for the corpus:
A = tfidf.fit_transform(tokenized)
#find similarity of article0 and article4
result = cosine_similarity(A[0], A[4]) #result is one number
#find similarity of article0 and all articles
result = cosine_similarity(A[0], A) #result is a numpy array with similarity scores for all articles
#Now you can sort your array to take highest 5
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python