Introduction

In previous Stobbs studies, we have outlined how relevance keywords can be used for filtering down large sets of candidate webpages into the subsets most likely to be relevant to a particular issue of interest. In general, the approach involves looking on each webpage for instances of the keywords in close proximity to the name of the brand of interest, and calculating a score for the page, based on the numbers of such pairs and the proximity of the terms to each other in each case. The pages with the highest score are then the most likely to be of interest[1]. This methodology is based on the technique used to calculate brand sentiment (which utilises sentiment keywords, but is otherwise essentially identical)[2].

The ability to filter in this way is an essential component of any brand-monitoring solution, allowing the most relevant pages to be quickly and efficiently identified, and reducing the requirement for expensive analyst resource to manually review large volumes of results.

In this study, we apply the approach to the identification of e-commerce websites likely to be associated with the sale of counterfeit or otherwise infringing goods, from a pool of larger product-related sites, based on identification terminology typically used by infringing sellers. This is of key importance for any brand-protection programme aiming to tackle the most egregious infringements, where counterfeits often present the greatest risk to revenue, and are typically the most readily enforceable.

The methodology utilises a list of ‘high-risk’ e-commerce keywords, which have previously been identified as being highly diagnostic of non-legitimate products when used in conjunction with a brand name. Examples include ‘dupe’, ‘mirror-quality’, ‘A-grade’, and so on. These terms are most usually applied to fashion and luxury products, but can be applicable to other areas.

In this study, we focus on general Internet content (i.e. ‘standalone’ e-commerce websites identified through search-engine queries), rather than on marketplaces specifically. This is partly because issues with general Internet content are, in general, the hardest to solve (because of the greater difficulty in finding relevant content, and the limitless range of ways in which product information can be presented on the page), but also because some marketplaces explicitly prohibit the posting of listings with branded product references (meaning many sellers will resort to use of brand misspellings and variations). Accordingly, an approach able to handle the analysis of general content can usually be applied to more well-defined channels (though with appropriate modifications, where necessary).

Findings

Proof-of-concept 1: luxury brands

In the first proof-of-concept, we consider the online sale of luxury-branded products, using Gucci and Chanel as representative brand examples. In each case, we generate a ‘pool’ of candidate pages for analysis by collecting the results returned by search engines (the first page of Google.com results, in this case) in response to query-terms consisting of the brand name combined with product-related keywords (‘handbag’, ‘shoes’, etc.) and e-commerce-related (and especially ‘high risk’) keywords (such as ‘buy’, ‘shop’, ‘cheap’, ‘discount’, ‘replica’, etc.). For each brand, this yielded just under 2,000 unique webpages for analysis. Of the pages which were accessible for analysis by our automated tool, the datasets produced 252 results yielding non-zero (positive) scores for Gucci, and 222 for Chanel. The assertion is that these pages are the ones most likely to be offering the sale of counterfeit products, based on the use of ‘high-risk’ keywords near to the brand names (and, moreover, that the pages with the higher scores will, in general, feature greater numbers of such references and/or brand/keyword pairs with closer proximity). These would therefore be the priority candidates for further analysis and potential enforcement.

Manual inspection of the results suggests that this is indeed the case, and almost all of the non-zero-scored pages feature what we would consider to be content of potential concern (Appendix A; Figures 1 and 2).

Figure 1: Examples of ‘high-risk’ pages for Gucci

Figure 2: Examples of ‘high-risk’ pages for Chanel

It is also worth noting that it may be necessary to ‘tune’ the keyword lists in response to the types of content being found and the terminology used. An earlier iteration using the ‘inspired’ as a ‘high-risk’ keyword (e.g. in the context of ‘Gucci-inspired’ items) was found to lead to a less ‘clean’ categorisation of results, due the more generic nature of this term and its tendency also to be used in other contexts (e.g. “It was not just the high quality clothes of the wealthy guests that inspired Gucci” and “During the 1930s Gucci became inspired by horse racing”[3]).

Proof-of-concept 2: electronics brands

The second case study considers electronics brands (using Samsung and Panasonic as representative examples, with appropriate product-related keywords – ‘tv’, ‘smartphone’, ‘headphones’, etc.), as an area where the sale of counterfeits is of particular concern because of the obvious safety implications.

In this case it was found that the keyword ‘discount’ (which was actually relatively diagnostic of ‘high-risk’ content for luxury brands) was less diagnostic for electronics, in part due to the large number of sites offering discount codes. This keyword was therefore removed from the list for this second analysis.

However, the approach was again effective, though a manual inspection of the positively-scored results found a smaller proportion of results of real potential concern than for the luxury brands, where the keywords are more directly applicable to infringing items. However, as in the first study, significant numbers of results of interest were successfully identified, and effectively prioritised using the keyword / scoring approach (Figures 3 and 4).

Figure 3: Examples of ‘high-risk’ pages for Samsung

Figure 4: Examples of ‘high-risk’ pages for Panasonic

Conclusion

Overall, the use of ‘high-risk’ e-commerce keywords in an analysis of proximity to brand terms has been found to be an effective way of prioritising sets of ‘candidate’ pages, in order to identify the subset which are most likely to be associated with the sale of counterfeits (even if the exact list may need to be ‘tuned’ to be most applicable to specific industry or product areas). The approach is therefore an important consideration in the deployment of automated brand-monitoring tools, to build efficiency into the analysis process and reduce the incidence of ‘false positives’.

In the specific cases of the live infringements identified in this analysis, the fact that all (by definition) have been live for long enough to have been indexed by search engines, and are highly ranked in response to relevant query terms, should be a cause for concern for the respective brand owners. Assuming that these brands do have brand protection services in place, the analysis highlights that (depending on the methodologies used) in many cases, relevant results may be being missed, are not effectively being identified from within wider datasets of identified findings, or that enforcement attempts may have been ineffective (which may particularly be the case for some of the more problematic jurisdictions, such as with .ru and .cn domains) – or some combination of all of the above points. This conclusion highlights the requirement for an effective programme for detection, prioritisation, and takedown to be employed by brand owners.

Appendix A: The most highly scored pages for luxury brands

Table A.1: Webpage titles for all results assigned a potential relevance/risk score of 300 or greater for Gucci

Table A.2: Webpage titles for all results assigned a potential relevance/risk score of 300 or greater for Chanel

[1] ‘Utilisation of relevance keywords for prioritising results in brand monitoring’, forthcoming Stobbs e-book

[2] https://www.iamstobbs.com/online-brand-prominence-and-sentimentebook

[3] https://thevintagecompactshop.com/blogs/antique-and-collectible-history/gucci-a-brief-history