Lade Inhalt...

Recommendation of products using the math-hypot function

von Krina Joshi (Autor) Dharni Patel (Autor)

Studienarbeit 2016 6 Seiten

Informatik - Angewandte Informatik


Recommendation Of Products Using Math-Hypot

Krina Joshi[1], Dharni Patel[2]

[1],[2] Department Of Computer Science, Rollwala Computer Center, Gujarat University

Abstract: Recommender systems have become an essential part of our day-to-day lives, when it comes to dealing with the overwhelming amount of information available, especially online. Recommender systems improve user experience and increase revenue in the context of online retail stores (Amazon, eBay), online news providers (Google News, BBC) and much more. Three different recommender system approaches namely Collaborative filtering (CF), Content-based filtering, Hybrid recommender systems are used on different recommendation websites. Problems faced by current recommendation and prediction systems are a cold start, scalability, accuracy, and sparsity. We aim to develop a recommendation system which would be more accurate, scalable and would improve performance when data are sparse.

1. Introduction

Recommendation framework is a data separating method, which furnishes clients with data, which he/she might be keen on.

Recommender frameworks have turned out to be to a great degree regular as of late, and are connected in an assortment of uses. The most mainstream ones are likely motion pictures, music, news, books, look into articles, seek inquiries, social labels, and items all in all. In any case, there are likewise recommender frameworks for specialists, partners, jokes, eateries, monetary administrations, life coverage, people (web based dating), and Twitter devotees.

Recommender frameworks commonly deliver a rundown of suggestions in one of two routes - through cooperative or substance based separating. Community oriented separating approaches building a model from a client's past conduct (things beforehand obtained or chose and additionally numerical appraisals given to those things) and additionally comparable choices made by different clients. This model is then used to anticipate things (or evaluations for things) that the client may have an enthusiasm for. Content-based sifting approaches use a progression of discrete qualities of a thing keeping in mind the end goal to prescribe extra things with comparative properties. These methodologies are frequently consolidated (see Hybrid Recommender Systems).

The contrasts amongst synergistic and substance based separating can be exhibited by looking at two prominent music recommender frameworks - and Pandora Radio. makes a "station" of prescribed melodies by watching what groups and individual tracks the client have listened to all the time and contrasting those against the listening conduct of different clients. will play tracks that don't show up in the client's library, however, are regularly played by different clients with comparative interests. As this approach influences the conduct of clients, it is a case of a collective sifting strategy.

Pandora utilizes the properties of a melody or craftsman (a subset of the 400 traits gave by the Music Genome Project) so as to seed a "station" that plays music with comparable properties. Client criticism is utilized to refine the station's outcomes, deemphasizing certain qualities when a client "abhorrence’s" a specific melody and stressing different traits when a client "likes" a tune. This is a case of a substance based approach.

Every kind of framework has its own particular qualities and shortcomings. In the above case, requires a lot of data on a client keeping in mind the end goal to make precise suggestions. This is a case of the frosty begin issue and is basic in community separating frameworks. While Pandora needs almost no data to begin, it is significantly more constrained in degree (for instance, it can just make suggestions that are like the first seed).

Recommender frameworks are a valuable other option to pursuit calculations since they help clients find things they won't have found independent from anyone else.

2. The Problem Statement

There are many difficulties in proposal framework:

Frosty Starter. For new clients and additionally things, no data to influence. Inadequate Data. Thing audits or buys are not exceptionally regular. Versatility Issue. The greater the information gets, the more calculation is required.

3. Proposed Work

Predication/recommendation algorithm that is quick, efficient and scalable compared to what has been done so far. Recommendation Algorithm using real-time data. Our main focus would be on the product-based recommendation system.

4. Implementation Requirements

1. Python Core Packages:
1.1 Scipy
1.2 Numpy
1.3 Matplotlib
1.4 Ipython/Jupyter Notebook

2. Dataset
2.1 Movies Dataset

3. System Configuration:
3.1 RAM: 4GB
3.2 Processor: Intel Pentium processor T4300
3.3 Operating system: Windows 7/8/10
3.4 Python Version:2.7 or more

5. Steps Of Implementation

illustration not visible in this excerpt

Figure 1. Steps Of Implementation

For implementing the algorithm we have used python libraries. After installing python libraries the first thing we need is dataset on which algorithm can be implemented. After finding the dataset we need to parse the dataset into required format. After parsing/splitting the data we have applied the algorithms on the data.

At the end, we will get the time complexity and accuracy of the data

5.1 Comparison of math.hypot with other algorithms:

We are proposing that, if we use math_hypot function in collaborative filtering we will get a scalable and accurate output as compared to another algorithms.

Table 1. Comparison Of Math.Hypot With Other Algorithms

illustration not visible in this excerpt

6. Results Obtained

For movie dataset,we have taken Age , Gender, Location, Occupation, Year of release as parameters.

We were able to run for different set of user recommendation database like 10000, 20000, 30000, 40000 ,50000, 60000, 70000, 80000,90000,100000 users

We were able to run above different recommendation algorithms like pearson, euclidean,hypot, spearman_rank on above set of user db

After modifying algorithm using SciPy, NumPy, MatPlotlib ,Math.hypot ;following are the results we have obtained:

Table 2. Comparison Of Different Algorithms Depending on theTime Taken(seconds) for USER ID 254,without using parameters(considerations)

illustration not visible in this excerpt

For user number 254,

If the dataset is of 1-10000 users then the time execution of euclidean distance,spearman,pearson-corelation ,math_hypot is 0.064,0.098,0.053,0.027 seconds respectively.

Similarly when we calculate for 20000,30000,40000,50000,60000,70000,80000,90000,100000 users math hypot gives better output.

illustration not visible in this excerpt

Figure 2. Graphical Representation of comparison of different algorithms for user-id 254

Table 2. Comparison of algorithms based on accuracy for Userid 254:

User have given ratings to following movies:({'Horror': 14, 'Action': 7, 'Sci-Fi': 4, 'Comedy': 4, 'Adventure': 3, 'Thriller': 3, 'Fantasy': 2, 'Drama': 2, 'Crime': 1, "Children's": 1, 'War': 1})

illustration not visible in this excerpt

Table 3. Comparison of algorithms based on accuracy for Userid 307:

User have given ratings to following movies:({'Drama': 90, 'Comedy': 44, 'Thriller': 26, 'Romance': 24, 'War': 24, 'Action': 16})

illustration not visible in this excerpt

Table 4. Comparison of algorithms based on accuracy for Userid 507:

User have given ratings to following movies:({'Drama': 27, 'Comedy': 11, 'Romance': 7, 'Thriller': 4, 'Mystery': 3, 'Crime': 3})

illustration not visible in this excerpt

Table 5. Comparison of algorithms based on accuracy for Userid 1674:

User have given ratings to following movies:({'Comedy': 19, 'Drama': 8, 'Romance': 6, "Children's": 3, 'Sci-Fi': 1, 'Fantasy': 1, 'Animation': 1, 'Thriller': 1})

illustration not visible in this excerpt

Table 6. Comparison of algorithms based on accuracy for Userid 2891:

User have given ratings to following movies:({'Comedy': 27, 'Drama': 18, 'Thriller': 10, 'Romance': 8, 'Action': 7})

illustration not visible in this excerpt

7. Conclusion

To develop and demonstrate a methodology which provides scalable recommendations

To develop and demonstrate a methodology which provides accurate recommendations

We have proposed to use math_hypot function with collaborative filtering algorithm.

As far as scalability and accuracy were concerned, we have proved that math_hypot gave more scalable and accurate results as compared to existing algorithms which are used for recommendation and prediction.

8. References

[1] Schedl, Markus. "On the Influence of User Characteristics on Music Recommendation Algorithms."Advances in Information Retrieval. Springer International Publishing, 2015.
[2] Park C, Lee J, Park G, Hyun J. Development of reservation recommendation algorithms for charging electric vehicles in smart-grid cities. International Journal of Smart Home. 2014 Jan 1.
[3] Zheng, Yong. "Deviation-based and similarity-based contextual slim recommendation algorithms."Proceedings of the 8th ACM Conference on Recommender Systems. ACM, 2014.
[4] Loc Nguyen .“ Introduction to A Framework of E-commercial Recommendation Algorithms”. American Journal of Computer Science and Information Engineering (AJCSIE) 09/2015
[5] Nagrale , Amita M., and Amol P. Pande . "User Preferences-based Recommendation System for Services Using Map Reduce Approach for Big Data Applications." (2015).
[6] Javari, Amin, and Mahdi Jalili. "A probabilistic model to resolve diversity–accuracy challenge of recommendation systems."Knowledge and Information Systems (2014).
[7] Ashkan, Azin, et al. "Diversified utility maximization for recommendations."Poster Proceedings of the 8th ACM Conference on Recommender Systems. 2014.
[8] Piller, Frank, et al. "Strategic Capabilities of Mass Customization Based E-Commerce: Construct Development and Empirical Test."System Sciences (HICSS), 2014 47th Hawaii International Conference on. IEEE, 2014.
[9] Serrano, Emilio, et al. "Strategies for avoiding preference profiling in agent-based e-commerce environments."Applied intelligence 40.1 (2014).
[10] Gilinsky Jr, Armand, Elizabeth C. Thach, and Karen J. Thompson. "Connectivity & Communication: A Study of How Small Wine Businesses Use the Internet."Journal of Small Business Strategy 14.2 (2015).
[11] Jiang, Yuanchun, et al. "Redesigning promotion strategy for e-commerce competitiveness through pricing and recommendation."International Journal of Production Economics 167 (2015).
[12] Cao, Mukun, et al. "Automated negotiation for e-commerce decision making: A goal deliberated agent architecture for multi-strategy selection."Decision Support Systems 73 (2015).
[13] Zhao, Xiangyu, et al. "Improving Top-N Recommendation Performance Using Missing Data."Mathematical Problems in Engineering 2015 (2015).
[14] Lim, Daryl, Julian McAuley, and Gert Lanckriet. "Top-N Recommendation with Missing Implicit Feedback."Proceedings of the 9th ACM Conference on Recommender Systems. ACM, 2015.
[15] Verstrepen, Koen, and Bart Goethals. "Top-N Recommendation for Shared Accounts."Proceedings of the 9th ACM Conference on Recommender Systems. ACM, 2015.
[16] Zheng, Yong, Bamshad Mobasher, and Robin Burke. "Cslim: Contextual slim recommendation algorithms."Proceedings of the 8th ACM Conference on Recommender Systems. ACM, 2014.
[17] Kim, Yonghwan, and Hsuan-Ting Chen. "Social media and online political participation: The mediating role of exposure to cross-cutting and like-minded perspectives."Telematics and Informatics 33.2 (2016): 320-330.
[18] Zheng, Qianru, Chi-Kong Chan, and Horace HS Ip. "IURA: An Improved User-based Collaborative Filtering Method Based on Innovators."Proceedings of the International MultiConference of Engineers and Computer Scientists. Vol. 1. 2014.
[19] Rausch, Katharina, et al. "Exploring subspace clustering for recommendations."Proceedings of the 26th International Conference on Scientific and Statistical Database Management, 2014.
[20] Qiao, Zhi, et al. "Improving Collaborative Recommendation via Location-based User-item Subgroup."Procedia Computer Science 29 (2014).


549 KB




Titel: Recommendation of products using the math-hypot function