news-recommendation
This is a spring webflux server which implements TF-IDF based article recommendation engine.
Authentication and Authorization
Since recommendations require some sort of user profile. so the relevant reactive REST endpoints are secured using spring Webflux security. We will implement token-based authentication and authorization using JWT provider.
Persistent Datastore
mlab mongodb free tier instance.
Reactive Weblux
Reactive programming require you to pipe the results from producers to consumer. This way the intermediate and final results are materialized only if they are needed. Hence there is technically no thread waiting for io. see the example.
public Mono<ServerResponse> setPreference(ServerRequest request) {
System.out.println("Start setPreference()");
Authentication authentication = SecurityContextHolder.getContext().getAuthentication();
String username = (String) authentication.getPrincipal();
Mono<Article> item = request.bodyToMono(Article.class);
return item.flatMap(i -> { i.setV(RunCmd(i)); return Mono.just(i); })
.flatMap(objectId -> this.feedService.setPreference(objectId, username)
.flatMap(x -> ServerResponse.ok().contentType(APPLICATION_JSON).bodyValue(x))
.switchIfEmpty(ServerResponse.notFound().build()));
}
NLP
The TF-IDF vector based similarity is used in recommendations. It is a very simple similarity algorithm but works well is real use cases. Currently, max features is 100 due to free tier of mongodb(and there is no native support for vector dot).
Database ER representation
