1
$\begingroup$

I am trying to create a decision tree to predict new locations a user would like to visit based on previous locations they have liked.

Here is my problem, I have two data sets,

  • a user data set which stores the tags from locations they have previously liked and how many times these tags have been liked by the user:example user dataset

  • a location dataset for locations and their associated tags:example location dataset

Is it possible, using decision trees (or another ML algorithm -please specify) to suggest/predict for a user which locations they will 'like' based on the tags of photos they have previously liked?

I am fairly new to machine learning and I am really struggling to think of a way to train my datasets if this is possible. Does anyone have any ideas?

Thank you in advance.

$\endgroup$
  • $\begingroup$ Do you have information what user liked what location. Like sparse user-location matrix? Without the additional features (grass, house, ...) $\endgroup$ – Tomáš Přinda Nov 25 '18 at 10:15
  • $\begingroup$ @TomášPřinda Yes, I have this information $\endgroup$ – Bob K Nov 25 '18 at 12:17
0
$\begingroup$

1) One super simple suggestion (Vector space model in information retrieval):

  • Each user is represented as an (row) vector
  • Each location is represented as a (row) vector

Use as similarity measure the scalar product (cos-similarity). Suggest the locations with the highest similarity score.

for cos-similarity see, eg https://nlp.stanford.edu/IR-book/html/htmledition/dot-products-1.html

2) If you have labels, i.e. users-location pairs (users that like locations) you can use an architecture similar to a "Siamese neural network" (maybe without weight sharing) for finding the similarity between users and locations. The idea is to map the users and the locations in another feature space where the similarity is computed by the neural network. Without knowing much more about the data (sparsity, synonyms etc.) it is difficult to say more if that makes sense or not. for an overview to Siamese NN, see https://www.quora.com/What-are-Siamese-neural-networks-what-applications-are-they-good-for-and-why

$\endgroup$

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.