ger/test/ger/ger_recommendations_for_people_tests.coffee

Good Enough Recommendations (GER) is a collaborative filtering based recommendations engine built to be easy to use and integrate into your application.

ger_tests = (ESM) ->
  ns = global.default_namespace

  describe 'recommending for a person', ->

    it 'should recommend similar things', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','b', expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1',  actions: {view: 1}, filter_previous_actions: ['view']))
        .then((recs) ->
          recs = recs.recommendations
          recs.length.should.equal 1
          recs[0].thing.should.equal 'b'
        )


    it 'should not return a weight of NaN if person similarity is 0', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','x', created_at: today, expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', created_at: yesterday, expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1',  actions: {view: 1}, filter_previous_actions: ['view'], neighbourhood_search_size: 1))
        .then((recs) ->
          for r in recs.recommendations
            throw "BAD WEIGHT #{r.weight}" if not _.isFinite(r.weight)

          throw "BAD Confidence #{recommendations_object.confidence}" if not _.isFinite(recs.confidence)
        )

  describe 'time_until_expiry', ->
    it 'should not return recommendations that will expire within time_until_expiry seconds', ->
      one_hour = 60*60
      one_day = 24*one_hour
      a1day = moment().add(1, 'days').format()
      a2days = moment().add(2, 'days').format()
      a3days = moment().add(3, 'days').format()

      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a'),

          ger.event(ns, 'p2','view','a'),
          ger.event(ns, 'p2','buy','x', expires_at: a1day),
          ger.event(ns, 'p2','buy','y', expires_at: a2days),
          ger.event(ns, 'p2','buy','z', expires_at: a3days)
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', time_until_expiry: (one_day + one_hour), actions: {view: 1, buy: 1}))
        .then((recs) ->
          recs = recs.recommendations
          recs.length.should.equal 2
          sorted_recs = [recs[0].thing, recs[1].thing].sort()
          sorted_recs[0].should.equal 'y'
          sorted_recs[1].should.equal 'z'
        )

  describe "minimum_history_required", ->
    it "should not generate recommendations for events ", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','b', expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', minimum_history_required: 2, actions: {view: 1}))
        .then((recs) ->
          recs.recommendations.length.should.equal 0
          ger.recommendations_for_person(ns, 'p2', minimum_history_required: 2, actions: {view: 1})
        ).then((recs) ->
          recs.recommendations.length.should.equal 2
        )


  describe "joining multiple gers", ->
    it "similar recommendations should return same confidence", ->
      ns1 = 'ger_1'
      ns2 = 'ger_2'
      bb.all([
        init_ger(ESM, ns1),
        init_ger(ESM, ns2)
      ])
      .spread (ger1, ger2) ->
        bb.all([

          ger1.event(ns1, 'p1','view','a', expires_at: tomorrow),
          ger1.event(ns1, 'p2','view','a', expires_at: tomorrow),
          ger1.event(ns1, 'p2','buy','b', expires_at: tomorrow),

          ger2.event(ns2, 'p1','view','a', expires_at: tomorrow),
          ger2.event(ns2, 'p2','view','a', expires_at: tomorrow),
          ger2.event(ns2, 'p2','buy','b', expires_at: tomorrow),
        ])
        .then( -> bb.all([
            ger1.recommendations_for_person(ns1, 'p1', {neighbourhood_size: 2, neighbourhood_search_size: 4, actions: {view: 1}}),
            ger2.recommendations_for_person(ns2, 'p1', {neighbourhood_size: 4, neighbourhood_search_size: 8, actions: {view: 1}})
          ])
        )
        .spread((recs1, recs2) ->
          recs1.confidence.should.equal recs2.confidence
        )


  describe "confidence", ->

    it 'should return a confidence ', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','action1','a', expires_at: tomorrow),
          ger.event(ns, 'p2','action1','a', expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', actions: {action1: 1}))
        .then((similar_people) ->
          similar_people.confidence.should.exist
        )

    it 'should return a confidence of 0 not NaN', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','action1','a', expires_at: tomorrow)
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', actions: {action1: 1}))
        .then((similar_people) ->
          similar_people.confidence.should.equal 0
        )

    it "higher weighted recommendations should return greater confidence", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p1','view','b', expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','b', expires_at: tomorrow),
          ger.event(ns, 'p2','view','c', expires_at: tomorrow),

          ger.event(ns, 'p3','view','x', expires_at: tomorrow),
          ger.event(ns, 'p3','view','y', expires_at: tomorrow),
          ger.event(ns, 'p4','view','x', expires_at: tomorrow),
          ger.event(ns, 'p4','view','z', expires_at: tomorrow),
        ])
        .then(->
          bb.all([
            ger.recommendations_for_person(ns, 'p1', actions: {view: 1})
            ger.recommendations_for_person(ns, 'p3', actions: {view: 1})
          ])
        )
        .spread((recs1, recs2) ->
          recs1.confidence.should.greaterThan recs2.confidence
        )

    it "more similar people should return greater confidence", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', expires_at: tomorrow),

          ger.event(ns, 'p3','view','b', expires_at: tomorrow),
          ger.event(ns, 'p4','view','b', expires_at: tomorrow),
          ger.event(ns, 'p5','view','b', expires_at: tomorrow),
        ])
        .then(->
          bb.all([
            ger.recommendations_for_person(ns, 'p1', actions: {view: 1})
            ger.recommendations_for_person(ns, 'p3', actions: {view: 1})
          ])
        )
        .spread((recs1, recs2) ->

          recs2.confidence.should.greaterThan recs1.confidence
        )

    it "longer history should mean more confidence", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','a', expires_at: tomorrow),

          ger.event(ns, 'p3','view','x', expires_at: tomorrow),
          ger.event(ns, 'p3','view','b', expires_at: tomorrow),
          ger.event(ns, 'p4','view','x', expires_at: tomorrow),
          ger.event(ns, 'p4','view','b', expires_at: tomorrow),
        ])
        .then(->
          bb.all([
            ger.recommendations_for_person(ns, 'p1', actions: {view: 1})
            ger.recommendations_for_person(ns, 'p3', actions: {view: 1})
          ])
        )
        .spread((recs1, recs2) ->

          recs2.confidence.should.greaterThan recs1.confidence
        )

    it "should not return NaN as conifdence", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', actions: {view: 1}))
        .then((recs) ->
          recs.confidence.should.equal 0
        )

  describe "weights", ->
    it "weights should determine the order of the recommendations", ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a', expires_at: tomorrow),
          ger.event(ns, 'p1','buy','b', expires_at: tomorrow),

          ger.event(ns, 'p2','view','a', expires_at: tomorrow),
          ger.event(ns, 'p2','view','c', expires_at: tomorrow),

          ger.event(ns, 'p3','buy','b', expires_at: tomorrow),
          ger.event(ns, 'p3','buy','d', expires_at: tomorrow),
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', actions: {view: 1, buy: 1}, filter_previous_actions: ['buy', 'view'] ))
        .then((recs) ->
          item_weights = recs.recommendations
          item_weights.length.should.equal 2
          item_weights[0].weight.should.equal item_weights[1].weight

          ger.recommendations_for_person(ns, 'p1', actions: {view: 1, buy: 2}, filter_previous_actions: ['buy', 'view'])
        )
        .then((recs) ->
          item_weights = recs.recommendations
          item_weights[0].weight.should.be.greaterThan item_weights[1].weight
          item_weights[0].thing.should.equal 'd'
          item_weights[1].thing.should.equal 'c'
        )

    it 'should negative weights should reduce recommended item', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','likes','a'),
          ger.event(ns, 'p1','likes','b'),

          ger.event(ns, 'p2','likes','a'),
          ger.event(ns, 'p2','hates','b'),
          ger.event(ns, 'p2','likes','x', expires_at: tomorrow),

          ger.event(ns, 'p3','likes','a'),
          ger.event(ns, 'p3','likes','b'),
          ger.event(ns, 'p3','likes','y', expires_at: tomorrow),

        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', actions: {likes: 1, hates: -1}))
        .then((recs) ->

          item_weights = recs.recommendations
          item_weights.length.should.equal 2
          item_weights[0].thing.should.equal 'y'
          item_weights[1].thing.should.equal 'x'
          item_weights[1].weight.should.be.lessThan item_weights[0].weight
        )

  describe "person exploits,", ->
    it 'recommendations_per_neighbour should stop one persons recommendations eliminating the other recommendations', ->
      init_ger(ESM)
      .then (ger) ->
        bb.all([
          ger.event(ns, 'p1','view','a'),
          ger.event(ns, 'p1','view','b'),
          #p2 is closer to p1, but theie recommendation was 2 days ago. It should still be included
          ger.event(ns, 'p2','view','a'),
          ger.event(ns, 'p2','view','b'),
          ger.event(ns, 'p2','buy','x', created_at: moment().subtract(2, 'days').toDate(), expires_at: tomorrow),

          ger.event(ns, 'p3','view','a'),
          ger.event(ns, 'p3','buy','l', created_at: moment().subtract(3, 'hours').toDate(), expires_at: tomorrow),
          ger.event(ns, 'p3','buy','m', created_at: moment().subtract(2, 'hours').toDate(), expires_at: tomorrow),
          ger.event(ns, 'p3','buy','n', created_at: moment().subtract(1, 'hours').toDate(), expires_at: tomorrow)
        ])
        .then(-> ger.recommendations_for_person(ns, 'p1', recommendations_per_neighbour: 1, actions: {buy: 5, view: 1}))
        .then((recs) ->
          item_weights = recs.recommendations
          item_weights.length.should.equal 2
          item_weights[0].thing.should.equal 'x'
          item_weights[1].thing.should.equal 'n'
        )


    it "a single persons mass interaction should not outweigh 'real' interations", ->
      init_ger(ESM)
      .then (ger) ->
        events = []
        for x in [1..100]
          events.push ger.event(ns, "bad_person",'view','t1', expires_at: tomorrow)
          events.push ger.event(ns, "bad_person",'buy','t1', expires_at: tomorrow)

        bb.all(events)
        .then( ->
          bb.all([
            ger.event(ns, 'real_person', 'view', 't2', expires_at: tomorrow)
            ger.event(ns, 'real_person', 'buy', 't2', expires_at: tomorrow)
            ger.event(ns, 'person', 'view', 't1', expires_at: tomorrow)
            ger.event(ns, 'person', 'view', 't2', expires_at: tomorrow)
          ])
        )
        .then( ->
          ger.recommendations_for_person(ns, 'person', actions: {buy:1, view:1})
        )
        .then((recs) ->
          item_weights = recs.recommendations
          temp = {}
          (temp[tw.thing] = tw.weight for tw in item_weights)
          temp['t1'].should.equal temp['t2']
        )

module.exports = ger_tests;