CHAPTER 4 THE BRAND ALLIANCED-BASED CUSTOMER ENGAGEMENT
4.4 B RAND A LLIANCE - BASED S EARCH E NGINE O PTIMIZATION FOR C AMPAIGN M ODULE
4.4.2 Brand Alliance-based Campaign Keyword Suggestion
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
34
structure to optimize the search result of brand alliance-based campaign in terms of link and keyword. After gathering back the CEB result, we can then further optimize the future campaign through the campaign feedback.
4.4.1 Brand Alliance-based Campaign
As we develop the concept of brand partnership, we focus on co-branding, co-advertising, and cause-related marketing. And what we also find is that upon three relationships, it is important to have a brand alliance-based campaign to fulfill the relationship. We define brand alliance-based campaign as the campaign that involves one or more stakeholders and have a same goal, such as popularizing the youbike in Taipei. That’s why the government, Giant, and Easycard co-creating for a combo service hold a series of brand alliance-based campaign and hit a success of the world.
4.4.2 Brand Alliance-based Campaign Keyword Suggestion
After the choice of brand partners, focal business can then further engage their brand partners to co-create the brand alliance-based campaign description.
Through discussions and decide the brand alliance-based campaign description, one question then should be asked is that “What query will the user uses to search for this brand alliance-based campaign?” Because focal firm might not have previous experience for brand alliance-based campaign’s user query, we should then give the brand alliance-based campaign a more precise suggested keyword to reconstruct the brand alliance-based campaign to make it more searchable.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
35
Keyword Suggestion (Abhishek & Hosanagar, 2007; Chen & Yu, 2008) has been discussed for decades but there are still several problems to solve under brand alliance-based campaign situations. First, the past methods are more focused on generating huge amounts of keywords for marketers to select. For example, a popular keyword generation online tool WordTracker (www.wordtracker.com) expands their keyword suggestion base through parsing back the meta tag keywords of top returned results. But in our case, we need more than just expanding a lot of keywords but to give an optimized result for the marketers. Second, since the query space needs a large number of dimensions, it will be hard to reach global optimum but it’s sufficient to identify the suboptimal solutions instead.
To optimize the keyword selection process for higher ranking with related campaign keywords, we propose to use genetic algorithm to solve campaign keyword suggestion problems. According to Cecchini (2007), genetic algorithm can be applied to find the relative query with more variety and also get the local optimal solutions. We will describe the selection process as follow (See Figure 4.4, 4.5):
Step 1 Initialization: Random select query words from brand alliance-based campaign description Step 2 Fitness function: Calculate the fitness of each brand alliance-based campaign query set Step 3 Selection: Select brand alliance-based campaign query on their maximum of fitness Step 4 Crossover: Crossover with probability Pc
Step 5 Mutation: Mutation with probability Pm through mutation pool
Step 6 Iteration: Go to step 2 for 20 times to get optimum brand alliance-based campaign suggested keywords
Figure 4.4 Brand alliance-based campaign keyword steps
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
36
Step 1 Initialization: First target business need to give a brand alliance-based campaign description that describe detail information about the brand alliance-based campaign. We will initially extract keywords from brand alliance-based keyword description to be our initial query population set.
Step 2 Fitness Function: We then put our query (q) from step1 for searching through search engine and return the search results(R). We define fitness as maximization of the cosine similarity ( ) between brand alliance-based campaign description (j) with each search result (𝑟𝑖).
(1) Step 3 Selection: We will select through the popular fitness selection method roulette-wheel. The roulette-wheel method (Back, 1996) defines the probability of selection as the individual query’s fitness (fi) divided by sum of total queries fitness.
(2)
Step 4 Crossover: There is probability Pc that the queries need to be crossover to recombine as new queries and others will stay the same query.
Crossover defines as the new term from selected pair queries. The new query will be extracted a random N terms from the first query and then combine the extraction of rest terms from second query.
Step 5 Mutation: There is probability Pm that the term in queries will be replaced by another term from the mutation pool to increase variety. The
Fitness q = max
𝑟𝑖𝜖𝑅(𝜎(𝑗, 𝑟𝑖)
probability of selection = 𝑓𝑖 𝑓𝑗
𝑁𝑗 =1
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
37
mutation pool is brand alliance-based campaign description and expanded by the snippet of each query’s search result.
Step 6 Iteration: After crossover and mutation, the new generated query sets have to go through iterations to get optimal results.
Output: suggested keyword for keyword optimization
Figure 4.5 Brand alliance-based campaign keyword suggestion services In this part, we will continue to use GoGoA as an example (See Figure 4.6).
After deciding GoGoB as brand partners, GoGoA contacts with GoGoB and wants to try to hold brand alliance-based campaign to fulfill the brand partner relationship. GoGoA accompanies with GoGoB and then develops their own
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
38
brand alliance-based campaign descriptions and get the suggested keyword from the engagement site services. They then rephrasing their brand alliance-based campaign description and reach more target audiences.
Figure 4.6 Example of brand alliance-based campaign keyword service