A recommendation engine for your php application
15 likes9,707 views
This document provides an overview of building a recommendation engine for an online book shop using PredictionIO. It discusses recommendation systems and different types of algorithms like content-based filtering, collaborative filtering using user-user and item-item similarities, and model-based approaches. It also covers installing and using PredictionIO, modeling event data, building recommendation engines, and implementing the engine to provide book and ebook recommendations to users based on actions like views, likes and purchases.
![Item cold start
Cannot predict ratings for new item till
some similar users have rated it [No
problem for content-based]](https://image.slidesharecdn.com/arecommendationengineforyourphpapplication-170129095410/85/A-recommendation-engine-for-your-php-application-41-320.jpg)
![Cosine similarity
[3,5]
[2,7]
[0,0]](https://image.slidesharecdn.com/arecommendationengineforyourphpapplication-170129095410/85/A-recommendation-engine-for-your-php-application-57-320.jpg)
![$ pio app new MyApp1
[INFO] [App$] Initialized Event Store for this app ID: 1.
[INFO] [App$] Created new app:
[INFO] [App$] Name: MyApp1
[INFO] [App$] ID: 1
[INFO] [App$] Access Key:
3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F
$ pio eventserver](https://image.slidesharecdn.com/arecommendationengineforyourphpapplication-170129095410/85/A-recommendation-engine-for-your-php-application-70-320.jpg)
![engine.json
"datasource": {
"params" : {
"appName": “MyApp1”,
"eventNames": [“buy”, “view”]
}
},](https://image.slidesharecdn.com/arecommendationengineforyourphpapplication-170129095410/85/A-recommendation-engine-for-your-php-application-83-320.jpg)
A recommendation engine for your php application
- 1. A recommendation engine for your PHP apps
- 2. 1) Intro to recommender systems 2) PredictionIO 3) Case Study
- 3. Definition: a system that help people finding things when the process of finding what you need is challenging because you have a lot of choices/ alternatives
- 4. So… it’s a search engine!
- 5. Search Engines Document base is (almost) static Queries are dynamic
- 6. Search Engines Create an index analyzing the documents Calculate relevance for a query: tf*idf
- 7. Recommender systems Document base is growing (eg: Netflix) Query is static: find something I like
- 9. Domain: news, products, … Helps defining what can be suggested
- 10. Purpose: sales, information, education, build a community What is TripAdvisor purpose?
- 11. Personalization levels • Non personalized: best sellers • Demographic: age, location • Ephemeral: based on current activities • Persistent
- 12. Types of input • Explicit: ask user to rate something • Implicit: inferred from user behaviour
- 13. Output • Prediction: predicted rating, evaluation • Recommendations: suggestion list, top-n, offers, promotion • Filtering: email filters, news articles
- 14. A model for comparison
- 15. User: people with preference Items: subject of rating Rating: expression of opinion (Community: space where opinions makes sense)
- 17. Non-personalized
- 18. Best seller Most popular Trending Summary of community ratings: eg best hotel in town
- 21. Hotel
- 22. Visitor Hotel
- 23. Visitor Hotel
- 24. Hotel A Hotel B Hotel C John 3 5 Jane 3 Fred 1 0 Tom 4 AVG 3.5 3 0
- 25. Content based
- 26. User rate items We build a model of user preference Look for similar items based on the model
- 28. Action 0.7 Sci Fi 3.2 Vin Diesel 1.2 … … https://www.amazon.com/Relevant-Search-applications-Solr-Elasticsearch/dp/161729277X http://www.slideshare.net/treygrainger/building-a-real-time-solrpowered-recommendation-engine
- 30. Need to know items content User cold start: time to learn important features for the user What if user interest change? Lack of serendipity: accidentally discover something you like
- 32. No need to analyze (index) content Can capture more subtle things Serendipity
- 33. User-User Select people of my neighborhood with similar taste. If other people share my taste I want their opinion combined
- 34. E.T 2 4 Joe 2 2 3 ? 1 5 5 2 4 … Tom 3 3 2 4 1 User-User: which users have similar tastes?
- 35. E.T 2 4 Joe 2 2 3 ? 1 5 5 2 4 … Tom 3 3 2 4 1 User-User: which users have similar tastes?
- 36. Item-Item Find an items where I have expressed an opinion and look how other people felt about it. Precompute similarities between items
- 37. E.T 2 4 Joe 2 2 3 ? 1 5 5 2 4 … Tom 3 3 4 1 Item-Item: which item are similar?
- 39. Sparsity When recommending from a large item set, users will have rated only some of the items
- 40. User Cold start Not enough known about new user to decide who is similar
- 41. Item cold start Cannot predict ratings for new item till some similar users have rated it [No problem for content-based]
- 42. Scalability With millions of ratings, computations become slow
- 44. Express my opinions as a set of tastes Compact representation of the matrix with relevant features
- 45. Rogue One 1 3 5 Joe 1 2 3
- 46. An example
- 47. Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8 How similar are Joe and Tom? How similar are Joe and Bob?
- 48. Only consider items both users have rated For each item - Calculate difference in the users’ ratings - Take the average of this difference over the items Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8
- 49. Sim(Joe, Tom) = (|8-2| + |2-7| + |7-5|)/3 = 13/3 = 4.3 Sim(Alice, Bob) = (|5-7| + |4-1| + |4-3| + |7-8|)/4 = 7/4 = 1.75 Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8
- 50. Now we have a score or weight for each user
- 51. Recommend what similar user have rated highly To calculate rating of an item to recommend, give weight to each user’s recommendations based on how similar they are to you.
- 52. use entire matrix or use a K-nn algorithm: people who historically have the same tastes as me aggregate using weighted sum weights depends on similarity
- 53. Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8 How similar are Item1 and Item2? How similar are Item1 and Item3?
- 54. Only consider items both users have rated For each item - Calculate difference in ratings for the 2 items - Take the average of this difference over the users Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8
- 55. Sim(I1, I2) = (|8-1| + |5-4| + |7-1|)/3 = 14/3 = 4,6 Sim(I1, I3) = (|2-5| + |5-7| + |7-7|)/3 = 5/3 = 1,6 Item1 Item2 Item3 Item4 Item5 Joe 8 1 ? 2 7 Tom 2 ? 5 7 5 Alice 5 4 7 4 7 Bob 7 1 7 3 8
- 56. As user-user, use whole matrix or identify neighbors
- 58. Our domain
- 59. Domain: online book shop, both paper and digital Recommend titles, old and news - Who bought this also bought - You might like
- 61. PredictionIO
- 62. Under the Apache umbrella Based on solid open source stack Customizable templates engines SDK for PHP
- 65. Installation via source code http://predictionio.incubator.apache.org/ install/install-sourcecode/
- 66. You can choose storage mysql/postgres vs elasticsearch+hbase
- 67. The event server
- 69. Pattern: user -- action -- item User 1 purchased product X User 2 viewed product Y User 1 added product Z in the cart
- 70. $ pio app new MyApp1 [INFO] [App$] Initialized Event Store for this app ID: 1. [INFO] [App$] Created new app: [INFO] [App$] Name: MyApp1 [INFO] [App$] ID: 1 [INFO] [App$] Access Key: 3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F $ pio eventserver
- 71. Server runs on port 7070 by default $ curl -i -X GET http://localhost:7070 {“status":"alive"}
- 72. $ curl -i -X GET “http://localhost:7070/ events.json?accessKey=$ACCESS_KEY"
- 73. Events modeling what can/should we model? rate, like, buy, view, depending on the algorithm
- 77. $set , $unset and $delete _pio* are reserved
- 78. setUser($uid, array $properties=array(), $eventTime=null) unsetUser($uid, array $properties, $eventTime=null) deleteUser($uid, $eventTime=null) setItem($iid, array $properties=array(), $eventTime=null) unsetItem($iid, array $properties, $eventTime=null) deleteItem($iid, $eventTime=null) recordUserActionOnItem($event, $uid, $iid, array $properties=array(), $eventTime=null) createEvent(array $data) getEvent($eventId)
- 79. Engines
- 81. D.A.S.E Architecture Data Source and Preparation Algorithm Serving Evaluation
- 82. $ pio template get apache/ incubator-predictionio-template- recommender MyRecommendation $ cd MyRecommendation
- 83. engine.json "datasource": { "params" : { "appName": “MyApp1”, "eventNames": [“buy”, “view”] } },
- 84. $ pio build —verbose $ pio train $ pio deploy
- 89. Implementation
- 91. 2 kind of suggestions - who bought this also bought (recommendation) - you may like (similarities)
- 92. View Like (add to basket, add to wishlist) Conversion (buy) Recorded in batch
- 93. 4 engines 2 for books, 2 for ebooks (not needed now) Retrained every night with new data
- 94. recordLike($user, array $item) recordConversion($user, array $item) recordView($user, array $item) createUser($uid)
- 95. getRecommendation($uid, $itype, $n = self::N_SUGGESTION) getSimilarity($iid, $itype, $n = self::N_SUGGESTION)
- 97. user cold start/item cold start if we don’t get enough suggestion switch to non personalized (also for non logged users)
- 99. user cold start/item cold start if we don’t get enough suggestion switch to non personalized (best sellers)
- 100. Michele Orselli CTO@Ideato _orso_ micheleorselli / ideatosrl mo@ideato.it https://joind.in/talk/93d2d
- 101. Links • http://www.slideshare.net/NYCPredictiveAnalytics/building-a-recommendation- engine-an-example-of-a-product-recommendation-engine?next_slideshow=1 • https://www.coursera.org/learn/recommender-systems-introduction • http://actionml.com/ • https://github.com/grahamjenson/ger