社群網路上的智慧行銷

Intelligent Marketing on Social Networks

鄧雅文(Ya-Wen Teng)

指導教授 ： 陳銘憲

國立臺灣大學/電機資訊學院/電機工程學研究所/博士(2018年)

https://doi.org/10.6342/NTU201800574

摘要

隨著社群網站的興起，越來越多的人習慣於通過社群網站進行交流和共享，從而產生社群影響力的現象，這種現象提供了一個很好的行銷機會，並對現實社會中的行銷相當有助益。在本論文中，我們著重在三個社群網路行銷上的問題，分別是針對數量限制和多品級產品的收入最大化問題，以及個人化推廣之問題。

首先，我們考慮的收益最大化問題中，所行銷之商品具有數量限制，對於這個問題，我們不僅要選取適當的一小部分人作為種子客戶，還要制定商品的價格。為了解決收益最大化問題，我們首先提出一種策略搜索演算法，稱為PRUB演算法，它能夠找出最佳解；接下來，我們利用PRUB作為基本架構，設計了一種啟發式演算法PRUB+IF，以貪婪的方式獲得可行的解。此外，基於PRUB架構下，我們進一步設計的兩階段種子選擇模式的PRUB+ER演算法，以更有效地找出有效的解。我們的實驗在真實社群網絡結構上證明了PRUB+ER的效率以及PRUB、PRUB+IF和PRUB+ER的有效性。

接下來，我們著重在對多品級產品的行銷上，來自同公司的不同品級的產品（例如iPhone 8、iPhone 8 Plus和iPhone X）乃是同時具有競爭和促銷關係。首先基於IC和凹圖模型，我們提出了一種新的影響力傳遞模型MuG-IC（Multi-Grade IC），用於描述對於多品級產品的社群影響力傳遞現象。之後，我們在MuG-IC下討論收益最大化的問題，並通過設計一種新的演算法PS（Pricing-Seeding）來解決問題，PS演算法通過迭代方式調整所建議的定價和所選擇的種子客戶。最後，我們在真實的網路結構上模擬客戶的估價分布，該實驗表明了PS演算法之有效性。

最後，由於社群影響力被廣泛地討論，過去許多學者針對各種擴散模型進行了影響力最大化等的研究，然而，據我們所知，現有的研究都沒有將社會科學中廣泛存在的興趣強度與社群影響力之間的相互作用納入影響力傳遞模型。為了彌補此一差距，我們提出了能夠捕捉動態影響力的ID模型。在這個ID模型下，我們提出個人化行銷的問題，乃是增加目標客戶在一個議題上的興趣強度。具體而言，為了使減少行銷成本，我們設計了一種新穎的演算法ISES來搜索行銷策略中作為種子客戶，並盡量最小化種子客戶的數量。ISES演算法能夠通過採用回溯搜索和採用修剪策略來選擇種子客戶，找出具有成本效益的解決方案。在DBLP的真實資料中，我們以實驗證明了ISES的有效性。

關鍵字

社群網路 ； 社群影響力 ； 行銷 ； 影響力傳遞模型 ； 收益最大化

並列摘要

As the social networking websites arise, more and more people are used to communicating and sharing with each other through the social networks, which leads to the phenomena of social influences.
Such phenomena provide a great chance of marketing and are able to benefit the real world business applications.
In this dissertation, we are interested in three practical issues about marketing on social networks.
We addressed the revenue maximization problems with a quantity constraint and on the multi-grade product, respectively, and the personalized promotion.

First, we aim for maximizing the revenue by considering the quantity constraint on the promoted commodity.
For this problem, we not only identify a proper small group of individuals as seeds for promotion but also determine the pricing of the commodity.
To tackle the revenue maximization problem, we first introduce a strategic searching algorithm, referred to as Algorithm PRUB, which is able to derive the optimal solutions.
After that, we exploit PRUB as a framework and propose a heuristic, Algorithm PRUB+IF, for obtaining feasible solutions in a greedy manner.
Moreover, we further devise Algorithm PRUB+ER, which is under a 2-stage seed selection schema within the PRUB framework, to figure out effective solutions more efficiently.
Our experiments demonstrate the good efficiency of PRUB+ER as well as the great effectiveness of PRUB, PRUB+IF, and PRUB+ER on real social network structure.

Next, we are interested in the marketing of the multi-grade product, where the different grades of a product from a company, such as iPhone 8, iPhone 8 Plus, and iPhone X, have both competitive and promotional relationships.
For the study, a new diffusion model named MuG-IC (Multi-Grade IC) is first proposed based on the IC and the concave graph models to describe the phenomena of social influences regarding the multi-grade product.
Afterwards, we then study the revenue maximization upon the MuG-IC and solve the problem by designing a novel algorithm named PS (Pricing-Seeding).
The PS algorithm can give proper suggestions of pricing each grade of the product and seeding customers by tuning the suggestions in an iterative manner.
The experiments conducted on the real network structure with simulated valuation distributions from Amazon.com demonstrate the effectiveness of the proposed algorithm.

Finally, for the widespread utilization of social influences, a lot of works such as influence maximization and innovation promotion have been studied on various diffusion models.
However, to the best of our knowledge, none of the existing works has incorporated the interplay between the intensity of interest and influence strength, which has been widely observed in social sciences, into the diffusion model.
To fulfill this gap, we propose the ID model that is able to capture the dynamic influence strength owing to the interplay.
Under this ID model, we address the novel utilization of dynamic influence strength for personalized promotion to grow the intensity of a target individual's interest in an issue.
In particular, to have the cost of promotion minimized, we introduce a novel Algorithm ISES to search for the least number of individuals as seeds in the promotion strategy.
The ISES algorithm is able to identify the cost-effective solution by adopting the backtracking search and employing pruning strategies.
On the real dataset of DBLP, the experiments demonstrate the effectiveness of ISES.

並列關鍵字

social networks ； social influences ； marketing ； influence diffusion model ； revenue maximization

參考文獻

[1] S. E. Asch. Effects of Group Pressure on the Modification and Distortion of Judgments. Carnegie Press, 1951.

[2] D. Easley and J. Kleinberg. Information cascades. In Networks, Crowds, and Markets: Reasoning about a Highly Connected World. Cambridge University Press, 2010.

[3] H. Fromlet. Predictability of financial crises: lessons from sweden for other countries. Business Economics, 2012.

[5] A. Sundararajan. Local network effects and complex network structure. The BE Journal of Theoretical Economics, 2007.

[6] M. Granovetter. Threshold models of collective behavior. American journal of sociology, 1978.